chore: triage maintenance review and close completed stories

.claude/memory/MEMORY.md

@@ -3,7 +3,10 @@
 - [Product selling points](product-selling-points.md) — key differentiators and landing page angles for neuron-tai
 - [User profile](user-profile.md) — who Dobromir is and how to work with him
 - [Project status](project-status.md) — 35/35 stories done; alpha hardening next
-- **Alpha hardening** — `.scratch/alpha-hardening/` (22 issues, ADRs 0016–0019, [README](../.scratch/alpha-hardening/README.md), [handoff](../.scratch/alpha-hardening/handoff.md))
+- **Alpha hardening** — `.scratch/alpha-hardening/` (22 issues, ADRs 0016–0019, [README](../../.scratch/alpha-hardening/README.md), [handoff](../../.scratch/alpha-hardening/handoff.md))
 - [Alpha hardening navigation](alpha-hardening-navigation.md) — locked fraud/auth decisions, Bucket-1 order, handoff pointers
-- **Node capability admission** — `.scratch/node-capability-admission/` (P0 plan: generic doctor/real-forward validation, fail-closed readiness, tracker admission gate; [PRD](../.scratch/node-capability-admission/PRD.md), [README](../.scratch/node-capability-admission/README.md), ADR-0023)
+- **Node capability admission** — `.scratch/node-capability-admission/` (P0 plan: generic doctor/real-forward validation, fail-closed readiness, tracker admission gate; [PRD](../../.scratch/node-capability-admission/PRD.md), [README](../../.scratch/node-capability-admission/README.md), ADR-0023)
 - **Distributed relay performance** — relay `/rpc` requester sockets are persistent per Route Session and Activation Seam as of 2026-07-10; `request_id` remains unique per activation while `X-Meshnet-Session` remains stable for KV state. Next low-risk priorities: persistent direct/loopback HTTP, seam byte/latency telemetry, then trace-driven zstd tuning.
+- **Distributed GGUF direction** — benchmark-gated native runtime: compare controlled Transformers/safetensors and whole-model llama.cpp lanes before expensive work; ship only for measured speed or model-fit advantage. Public parallelism is contiguous Shards in an Inference Route; concurrency comes from per-node continuous batching across isolated Route Sessions, while tensor/expert collectives stay inside optional trusted composite providers. Native data plane uses versioned Protobuf over long-lived gRPC/HTTP2 seam streams, with existing relay carrying the same opaque frames when needed. llama.cpp/GGML remains the substrate behind a project-owned standalone worker and small pinned fork; vLLM is an optional complete managed provider and concept donor, not a fork. Nakshatra, `prima.cpp`, `llama-gguf`, LiGGUF and historical GPUStack are source/test donors only. Active plan: [README](../../.scratch/distributed-gguf-runtime/README.md), [architecture](../../.scratch/distributed-gguf-runtime/architecture.md), [PRD](../../.scratch/distributed-gguf-runtime/PRD.md), [Ralph backlog](../../.scratch/distributed-gguf-runtime/prd.json). Research: [landscape](../../docs/research/distributed-gguf-landscape.md), [GitHub follow-up](../../docs/research/distributed-gguf-github-followup.md), [vLLM](../../docs/research/vllm-distributed-gguf-assessment.md).
+- [DGR ROCm setup](dgr-rocm-setup.md) — version-matched TheRock SDK layout, relocated devel payload, verified `gfx1151` HIP llama.cpp build, and GPU-diagnostic boundary.
+- **DGR-004 llama.cpp boundary** — `packages/node/native/llama/` locks `e920c523e3b8a0163fe498af5bf90df35ff51d25`, with a one-patch CMake marker and fail-closed clean materialize/apply/build/smoke harness. This is infrastructure only; stock GLM dense fallback remains uncertified.

.claude/memory/dgr-rocm-setup.md

@@ -0,0 +1,29 @@
+# DGR ROCm and llama.cpp setup
+
+As of 2026-07-13:
+
+- Project ROCm runtime: `/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv-rocm`
+- ROCm/TheRock build: `7.13.0a20260513`, target `gfx1151`
+- `rocm-sdk-devel` is installed. Its expanded SDK lives under the venv at
+  `site-packages/_rocm_sdk_devel`.
+- The wheel's redundant packaged payload was relocated to
+  `/home/popov/.local/share/rocm-sdk/7.13.0a20260513/rocm_sdk_devel` and symlinked
+  back into the venv because installing both packaged and expanded forms filled
+  the mounted drive. Do not reinstall it blindly; the wheel expands beyond
+  20 GB.
+- HIP llama.cpp source: `/run/media/popov/d/DEV/llamacpp/llama.cpp`, commit
+  `e920c523e3b8a0163fe498af5bf90df35ff51d25` (version 9991).
+- HIP build: `/run/media/popov/d/DEV/llamacpp/llama.cpp/build-hip`
+- HIP `llama-server` SHA-256:
+  `b6bb4da687dbde86e243ba006cef05919b7b97255cd7e2371e1d451220aca139`
+- Verified device: `ROCm0: Radeon 8060S Graphics`, `gfx1151`.
+- Model artifacts remain under `/run/media/popov/DATA/llm`; none were put under
+  `/home`.
+
+DGR-001's immutable contract remains CPU-only. GPU evidence uses the distinct
+signed `gpu-diagnostic` profile because llama-server process VRAM is not yet
+measurable by the benchmark driver. The profile must capture measured
+llama-server startup evidence for `ROCm0` and the actual offloaded/total layer
+count; configured `device` and `n_gpu_layers` values alone are not evidence.
+The accepted signer fingerprint is anchored in
+`.scratch/distributed-gguf-runtime/trusted-evidence-signers.json`.

.scratch/distributed-gguf-runtime/ADR-0020-distributed-gguf-runtime.md

@@ -1,177 +1,121 @@
-# ADR-0020: Distributed GGUF/llama.cpp Runtime With Per-Shard Local KV
+# ADR-0020: Lean Native Distributed GGUF Runtime
 
-Status: Proposed
+Status: Accepted
+Date: 2026-07-13
 
 ## Context
 
-The project currently uses PyTorch/Transformers for real model shards. That decision was captured in ADR-0001 because llama.cpp RPC at the time required the primary node to load the full model and distribute weights to workers, which conflicted with the desired model where nodes independently hold shards.
+The project currently uses Transformers/safetensors as its real model execution backend. This provides broad architecture coverage and a correctness reference, but reported and observed consumer CPU/GPU inference performance motivates evaluating llama.cpp/GGML and quantized GGUF.
 
-We now want to serve very large open models, including GLM-5.2 and Ornith-class MoE models, over a torrent-like inference marketplace. CPU and mixed consumer hardware matter. LM Studio and llama.cpp demonstrate much better CPU/GGUF performance than our current PyTorch CPU path. The user also has a personal relationship with Georgi Gerganov, making upstream collaboration plausible.
+The product objective is not merely local GGUF serving. It is performant concurrent inference for top open models whose weights do not fit on one consumer node. The project already owns the Tracker, Inference Route, Route Session, Activation Seam, local Hot KV State, relay/direct transport, cancellation, telemetry, billing, and capability admission.
 
-The current distributed PyTorch path is not yet production-grade: it recomputes the full growing sequence for every output token and disables KV cache inside manual layer calls. It sends hidden activations across seams, not KV, but those activations currently cover the full sequence every decode step.
+Research audited llama.cpp RPC, GPUStack/llama-box, Nakshatra, prima.cpp, llama-gguf, LiGGUF, vLLM and its GGUF plugin, Petals, exo, and related projects. No repository provides the complete public-network contract. llama.cpp is the strongest GGUF execution substrate. vLLM has mature managed-cluster parallelism and scheduling concepts but its PP/TP/EP runtime assumes a static trusted distributed world and is unsuitable as the public Shard runtime.
+
+The project must remain lean and avoid combining several half-integrated inference control planes.
 
 ## Decision
 
-Adopt a distributed GGUF/llama.cpp runtime track while keeping PyTorch as the reference and fast-architecture backend.
+### Primary native runtime
 
-The runtime model is:
+Use llama.cpp/GGML through one standalone C++ Shard worker and a small exact-commit patch stack.
 
-- GGUF/model artifacts are distributed through torrent/content-addressed storage.
-- Nodes independently acquire and verify artifacts; no root node streams model weights to workers at session start.
-- Tracker chooses a sticky route covering all layers.
-- Each node owns hot KV/state for the layers it executes.
-- Prefill sends chunked activations through the route and builds local per-shard KV.
-- Decode sends one-step activations through the route and appends local KV at every shard.
-- Cache/CDN servers store cold artifacts and optional prefix/session snapshots, not hot per-token KV.
-- Context is capped at 128K for the first serious product path.
+The patch scope is limited to:
 
-## Technical Framework
+- Range-aware GGUF tensor ownership/loading.
+- Architecture-defined intermediate boundary input/output.
+- Intermediate output before tail normalization/head.
+- Layer-filtered KV and external session-to-sequence mapping.
 
-The design separates five planes:
+Meshnet networking, routing, admission, billing, telemetry, and work evidence stay outside llama.cpp.
 
-- **Control plane**: tracker registry, coverage map, route selection, session lifecycle, telemetry, billing, and audit.
-- **Artifact plane**: Shard Swarms, GGUF/safetensors/tokenizer files, manifests, hashes, and local node storage.
-- **Execution plane**: active Inference Route, chunked prefill, one-step decode, and hidden-state movement across activation seams.
-- **Session state plane**: per-shard Hot KV State on route nodes, plus optional Prefix Snapshots outside the hot loop.
-- **Economics/trust plane**: reward accounting, validation events, slash proofs, public/private route policy.
+Nakshatra, prima.cpp, llama-gguf, LiGGUF, and historical GPUStack are source/test donors only. Their repositories are not runtime dependencies.
 
-Hard invariants:
+### Distributed parallelism
 
-1. Public-network Shards are contiguous layer ranges.
-2. Hot KV State is local to the node serving that Shard in that Route Session.
-3. Artifact distribution and route execution are separate systems.
-4. Decode seam payload must be `O(hidden_size)`.
-5. Prefill may be `O(sequence_length * hidden_size)`, but only in bounded chunks.
-6. The tracker chooses routes; nodes do not negotiate route topology peer-to-peer.
-7. Model/backend-specific cache internals stay behind backend capability reports.
-8. PyTorch remains the correctness/reference backend while llama.cpp/GGUF becomes the performance backend.
-9. Streaming responses are preferred when feasible; Generation Telemetry is always required.
+The first public-network primitive is layer/pipeline parallelism through contiguous Shards in an Inference Route.
 
-The full challenge register is in [technical-challenges.md](./technical-challenges.md). The open decision gates are in [decision-framework.md](./decision-framework.md).
+Per-node continuous batching combines decode steps from compatible active Route Sessions. Multiple complete routes provide data parallelism.
 
-Resolved gate:
+Tensor and expert parallel collectives may later operate inside one trusted composite node or managed cluster represented as one provider. They are not public WAN routing primitives.
 
-- Public-network Shards are layer ranges. Tensor-parallel/ring execution belongs inside a trusted node, colocated pod, or future composite node abstraction, not as the v1 public routing primitive.
-- Hot KV State is local to each route node for the Shard it serves. Cache servers may store Prefix Snapshots, but they are not part of the per-token decode path.
-- Distributed Route Session and Hot KV State semantics will be proven in the PyTorch route before llama.cpp/GGUF is extended for layer-boundary execution.
-- Streaming responses are preferred when feasible. Realtime Generation Telemetry is required so clients can see phase, generated token count, and tokens/sec even during prefill or non-streaming fallback paths.
-- llama.cpp/GGUF work targets upstreamable `libllama`/ggml hooks. A prototype fork is acceptable for exploration, but a permanent fork is not the plan.
-- Model targeting is two-tiered: use a small llama.cpp-supported GGUF model for the first protocol smoke test, then use `deepseek-ai/DeepSeek-V4-Flash` as the first serious large-model target. GLM-5.2 and Ornith remain later support audits.
-- Alpha fails Route Sessions on route-node loss instead of attempting automatic route repair. Repair requires compatible Prefix Snapshots and is a later capability.
-- v1 activation transfer stays on binary HTTP as defined by ADR-0008. QUIC/WebRTC/custom transport can be introduced later behind the same activation protocol.
+### Transport
 
-## Non-Goals
+Use gRPC over HTTP/2 with Protocol Buffers for the native Python/C++ Shard data plane.
 
-- Do not put remote cache servers in the per-token hot KV path.
-- Do not require every node to hold the full model.
-- Do not fork llama.cpp long-term if upstream APIs can support the needed layer-boundary hooks.
-- Do not target GLM-5.2 or Ornith first; prove the route/KV protocol on a simpler well-supported GGUF model, then target DeepSeek-V4-Flash as the first serious large model.
+- One long-lived bidirectional stream per Route Session Activation Seam.
+- Deadlines, cancellation, flow control, TLS/authentication hooks, structured status, and generated schemas.
+- Bounded chunks for prefill and a small decode fast path.
+- Existing relay infrastructure may carry the same versioned protobuf frames as opaque binary when direct connectivity is unavailable.
+- OpenAI client APIs remain HTTP/SSE; existing Tracker APIs remain unchanged.
 
-## Options Considered
+The boundary payload is a versioned named-tensor bundle because architecture boundaries may require more than one tensor.
 
-### A. Keep PyTorch-only distributed inference
+### vLLM
 
-Pros:
+Do not fork vLLM for public distributed Shards and do not transplant PagedAttention, Torch process groups, or the vLLM GGUF plugin into the llama.cpp worker.
 
-- Easy access to new Hugging Face architectures.
-- Transformers has mature single-process KV semantics.
-- Existing code already loads shards.
+Allow unmodified vLLM as an optional whole-model backend or managed TP/PP/EP cluster represented as one logical provider.
 
-Cons:
+Adapt only small control-plane concepts:
 
-- CPU inference is much slower than llama.cpp/GGUF.
-- Current distributed path bypasses `generate()` and disables cache.
-- Quantized GGUF ecosystem and LM Studio users are outside the runtime.
+- Named intermediate bundles.
+- Continuous batching and request ownership.
+- Versioned cache-transfer compatibility fingerprints.
+- Explicit transfer failure/abort lifecycle.
+- Load telemetry and fair tie-breaking.
 
-### B. Use llama.cpp only as a full local model backend
+### Benchmark gate
 
-Pros:
+GGUF performance is a hypothesis. Before expensive native work, compare the current Transformers/safetensors recipe with whole-model llama.cpp on controlled model, hardware, prompt, context, output, sampling, concurrency, memory, and quality lanes.
 
-- Quick performance win for nodes with enough RAM/VRAM.
-- Minimal coordination with distributed protocol.
+Later distributed release gates use thresholds locked before implementation results are known. The native track stops if llama.cpp/GGUF offers neither a meaningful performance benefit nor a meaningful model-fit benefit at useful speed.
 
-Cons:
+### Concurrency
 
-- Does not unlock 397B/753B-class models for ordinary nodes.
-- Does not solve marketplace layer routing.
+A native worker must isolate `(Route Session ID, route epoch)` through a llama sequence or bounded context and must not serialize all generations behind one global serving sequence.
 
-### C. Distributed GGUF with per-shard local KV (chosen)
+The node admits sessions against weight/KV/scratch budgets, batches compatible decode steps, prevents prefill starvation, applies backpressure, and exposes queue/batch/KV telemetry.
 
-Pros:
+### Architecture certification
 
-- Aligns with torrent artifact distribution.
-- Avoids root streaming weights to workers.
-- Uses llama.cpp/GGUF performance where supported.
-- Compatible with public node rewards by layer/work contribution.
-- Scales KV memory by layer range.
+Dense Llama-family is first. Qwen3/Qwen3-MoE is a separate explicit adapter. Every architecture/backend/recipe remains registered-but-dark until a real distributed forward, parity test, concurrency test, and capability admission pass.
 
-Cons:
+## Alternatives rejected
 
-- Requires new runtime APIs around layer-boundary hidden states and per-session KV.
-- Requires model-specific cache metadata for DSA/MLA/hybrid attention.
-- Harder to debug than single-process `generate()`.
+### Fork vLLM for the public mesh
 
-### D. Centralized KV cache servers
+Rejected because extracting its PP/TP/EP stages requires replacing static process groups, rank lifecycle, scheduler, request ownership, cache layout, failure behavior, and hardware assumptions. This would create a large difficult fork while discarding much of vLLM's core architecture.
 
-Pros:
+### llama.cpp RPC as the public protocol
 
-- Easier apparent session failover.
-- Central accounting of active cache.
+Rejected because it exposes coordinator-owned raw GGML devices, not independent Shards. Its trust, security, failure, cache, and per-node accounting model is unsuitable for arbitrary volunteer nodes.
 
-Cons:
+### Adopt Nakshatra or prima.cpp wholesale
 
-- Puts remote storage in the per-token hot path.
-- Adds bandwidth and latency at the worst possible point.
-- Creates consistency and privacy problems.
+Rejected because their repositories, build reproducibility, session/concurrency semantics, architecture coverage, protocol identity, and control planes do not satisfy the project contract. Their partial-loading and boundary work remains valuable evidence.
 
-Rejected for hot decode. Accepted only for cold prefix snapshots and failover checkpoints.
+### Build a custom GGUF engine
+
+Rejected because llama.cpp already provides the parser, kernels, architecture graphs, KV, tokenizer, and heterogeneous backends. Reimplementing these would spread effort and increase correctness risk.
+
+### Invent a custom transport
+
+Rejected. gRPC/HTTP2 already provides mature streaming, flow control, deadlines, cancellation, TLS, and cross-language schema generation.
 
 ## Consequences
 
-- ADR-0001 should eventually be amended: PyTorch remains valid, but llama.cpp/GGUF becomes a first-class backend.
-- The activation protocol must split prefill and decode explicitly.
-- Session IDs must be stable across the full request. The current fresh UUID-per-hop-call behavior must change.
-- Backends must report cache budget and cache compatibility.
-- Tracker route selection must include disk, memory pressure, cache warmth, and network latency.
-- Billing can be based on layer work, prefill tokens, decode tokens, and observed route participation.
-- Client UX should stream token deltas when feasible and must include route-session progress telemetry even when token deltas are not streamed.
+- The critical path contains Meshnet, one standalone worker, and one small pinned llama.cpp patch stack.
+- Transformers/safetensors remains the correctness reference and fallback for unsupported architectures.
+- Whole-model llama.cpp and vLLM managed clusters remain useful optional provider types.
+- The first milestone emphasizes controlled benchmark, parity, concurrent KV, and real two-machine evidence rather than a large-model demo.
+- Upstream collaboration with llama.cpp targets generic local hooks only; the project remains able to ship a narrow pinned fork if upstream acceptance takes time.
+- QUIC, public tensor parallelism, disaggregated prefill, speculative decode, route repair, and KV migration remain deferred until the core route passes release gates.
 
-## Required Runtime Capabilities
+## Verification gates
 
-PyTorch path:
-
-- manual layer calls with `past_key_values` / model-specific cache object
-- per-shard session cache store
-- prefill chunk append
-- decode step append
-- stable session lifecycle endpoints
-
-llama.cpp/GGUF path:
-
-- full local GGUF serving
-- layer/tensor map extraction from GGUF
-- optional partial layer loading or mmap-backed selected execution
-- inbound hidden-state execution from arbitrary start layer
-- outbound hidden-state return at stop layer
-- per-session KV ownership for loaded layers
-- cache budget/compatibility introspection
-- GLM-5.2 DSA support when upstream/runtime supports it
-
-## Implementation Plan
-
-1. Add full-model `LlamaCppBackend` using `llama-server` or `libllama`.
-2. Implement distributed KV in the PyTorch path to prove semantics.
-3. Add session lifecycle and prefill/decode wire protocol.
-4. Add model artifact manifest and torrent seeding metadata.
-5. Prototype localhost two-process llama.cpp layer boundary execution.
-6. Generalize to network route.
-7. Bring in GLM-5.2/Ornith once backend support and cache accounting are verified.
-
-## Acceptance Criteria
-
-- A two-node localhost route can prefill once and decode N tokens without recomputing the full prompt.
-- Seam payload during decode is `O(hidden_size)`, not `O(sequence_length * hidden_size)`.
-- Per-node KV memory grows with owned layer count and context length.
-- Route loss during alpha fails cleanly with explicit reason.
-- Full local GGUF backend outperforms PyTorch CPU on a supported model.
-- Artifact manifest can identify exactly which files/chunks a node must seed for its advertised layer range.
+1. Controlled safetensors-versus-GGUF performance contract.
+2. Two-process local range parity.
+3. Four-session concurrent KV isolation.
+4. Real two-machine execution using both Shards.
+5. End-to-end performance/fit advantage over the current distributed route.
+6. Separate Qwen3-family architecture certification.

.scratch/distributed-gguf-runtime/GLM-5.2-MAX-ALPHA-ROADMAP.md

@@ -0,0 +1,384 @@
+# GLM-5.2 Max distributed alpha roadmap
+
+Status: proposed executable epic target
+Last updated: 2026-07-13
+
+## Executive decision
+
+The alpha-release target is the exact open-weight model `zai-org/GLM-5.2` served with `reasoning_effort=max` from the smallest published Unsloth GGUF recipe, `UD-IQ1_S`, across multiple consumer machines.
+
+“Max” is a reasoning mode selected by the chat template/API request. It is not a separate model checkpoint.
+
+Alpha is earned only when the real target model:
+
+1. cannot fit within any one participating node's admitted memory;
+2. loads as contiguous layer Shards on at least two physical consumer machines;
+3. performs real GLM-5.2 MoE + DSA + IndexShare computation on every selected node;
+4. responds through the existing OpenAI-compatible Meshnet API with `reasoning_effort=max`;
+5. passes locked parity, usefulness, performance, telemetry, cancellation, and cleanup checks; and
+6. stores all model artifacts on configured mounted-drive storage, never under `/home`.
+
+The shortest safe path is not “support every GGUF architecture.” Dense Llama remains a small structural fixture. GLM-5.2 moves onto the critical path immediately after exact recipe identity and the pinned llama.cpp boundary. Qwen expansion, 1M context certification, MTP/speculative decoding, broad concurrency, and automatic route repair are post-alpha.
+
+## 1. Exact target contract
+
+### 1.1 Source model
+
+| Field | Locked target |
+|---|---|
+| Official repository | `zai-org/GLM-5.2` |
+| Official revision observed 2026-07-13 | `b4734de4facf877f85769a911abafc5283eab3d9` |
+| Model-weight license | MIT |
+| Official code/documentation license | Apache-2.0 |
+| Architecture | `glm_moe_dsa` / `GlmMoeDsaForCausalLM` |
+| Official architecture label | 744B total / approximately 40B active per token |
+| Exact stored checkpoint tensors | 753,329,940,480 parameters |
+| Transformer layers | 78 backbone layers plus one shared NextN/MTP layer in the artifact |
+| Layer types | first 3 dense; remaining 75 sparse MoE |
+| Routed experts | 256 |
+| Experts selected | 8 routed experts plus shared expert path |
+| Hidden width | 6,144 |
+| Attention | MLA under DSA, lightning indexer top-k 2,048 |
+| IndexShare | indexer roles are encoded by `indexer_types`; consumers reuse prior Full-layer indices |
+| Architectural maximum context | 1,048,576 tokens |
+| Alpha reasoning mode | `reasoning_effort=max` |
+
+The runtime must derive these values from the pinned artifact and fail closed on contradictory metadata. Marketing names are not compatibility identity.
+
+### 1.2 Alpha GGUF artifact
+
+| Field | Locked target |
+|---|---|
+| GGUF repository | `unsloth/GLM-5.2-GGUF` |
+| GGUF revision observed 2026-07-13 | `abc55e72527792c6e77069c99b4cb7de16fa9f23` |
+| Quantization | `UD-IQ1_S` |
+| Files | six GGUF shards |
+| Exact published bytes | 216,715,360,960 bytes |
+| Binary GiB | 201.832 GiB |
+| Published quality indication | about 76.2% top-1 agreement with the high-precision reference on Unsloth's quantization analysis |
+| Mounted storage rule | configured mounted drive only; never `/home` |
+
+Before downloading 216.7 GB, DGR-017 must generate a checked-in target manifest containing repository revisions, expected filenames, byte sizes, and resolved LFS SHA-256 values. Download is resumable and verified before route admission.
+
+`UD-IQ1_M` (228,492,966,624 bytes / 212.801 GiB) is the first diagnostic fallback if `UD-IQ1_S` exposes a runtime or quality defect. It does not satisfy the explicit “lowest quantization” alpha target unless the target contract is changed by human review.
+
+### 1.3 Runtime semantics required for alpha
+
+Required:
+
+- GGUF parsing and quantized kernels from one exact llama.cpp pin.
+- Correct GLM-5.2 MoE routing, selected experts, and shared expert.
+- Correct compressed MLA KV cache for locally owned layers.
+- Native DSA lightning indexer and sparse attention.
+- Correct IndexShare Full/Shared role execution from artifact metadata.
+- Range-owned contiguous transformer layers; each owned layer keeps all of its experts local.
+- Head-only embeddings and tail-only final norm/output head/sampling.
+- Architecture-defined activation boundary and optional DSA index sideband.
+- `reasoning_effort=max` chat-template behavior through the public API.
+- F32 seam correctness lane and a separately certified production activation dtype.
+
+Not required for alpha:
+
+- MTP/speculative decoding. The trailing NextN tensors may be loaded or explicitly excluded according to a certified recipe, but cannot be silently misinterpreted.
+- Full 1,048,576-token context.
+- Continuous batching beyond one target session.
+- Public-WAN tensor or expert parallel collectives.
+- Automatic mid-generation repair or KV migration.
+- Every CPU/GPU backend combination.
+
+## 2. Minimum resource envelope
+
+### 2.1 Weight and runtime memory
+
+The smallest artifact occupies 201.832 GiB before KV, DSA indexer state, scratch buffers, backend workspaces, process memory, and the operating system. **224 GiB aggregate runtime-accessible memory is only the experimental hard-fit floor**, consistent with Unsloth's approximate 223 GB one-bit requirement. It is not a conservative operational envelope.
+
+For admission, each node reserves:
+
+```text
+max(20% of physically usable memory, 8 GiB)
+```
+
+The remainder is the combined weight-plus-KV placement budget. Actual peak scratch is measured by backend/context and can force one extra node. Unified memory is counted once: integrated-GPU “VRAM” must not be added again to the same physical system RAM.
+
+| Physical usable tier | Minimum reserve | Weight + KV placement budget | IQ1_S 16K arithmetic minimum | Operational position |
+|---:|---:|---:|---:|---|
+| 32 GiB | 8.0 GiB | 24.0 GiB | 9 nodes | use 10 if attempted; latency-heavy |
+| 48 GiB | 9.6 GiB | 38.4 GiB | 6 nodes | possible; latency-heavy |
+| 64 GiB | 12.8 GiB | 51.2 GiB | 4 nodes | hard minimum; **5 recommended** |
+| 96 GiB | 19.2 GiB | 76.8 GiB | 3 nodes | recommended |
+| 128 GiB unified/system | 25.6 GiB | 102.4 GiB | 2 nodes | arithmetic hard minimum; **3 recommended** |
+
+The planner must use exact tensor byte ownership, not equal percentages. Embeddings, final head, dense versus MoE layers, shared experts, indexer tensors, quant block alignment, KV distribution, and backend workspace make equal layer counts unequal in memory.
+
+Recommended first target route: **three 96/128-GiB-class physical machines** or **five 64-GiB-class machines**, on the same wired switch with mounted model storage. Four 64-GiB or two 128-GiB machines are fit probes only and qualify solely if exact placement and measured peak-memory evidence retain the required reserve with no swap/overcommit.
+
+### 2.2 KV cache
+
+GLM-5.2 MLA caches 576 latent/rope values per token per backbone layer. Correct DSA also caches 128-dimensional indexer keys: ideally only for the 21 Full indexer layers, while the current experimental implementation may allocate them across all 78 layers. Alpha locks **Q8_0 KV** for quality and budgets the conservative current-implementation layout.
+
+| Context × concurrency | MLA-only Q8 | Optimized DSA Q8 | Conservative current-DSA Q8 | Conservative current-DSA F16 |
+|---:|---:|---:|---:|---:|
+| 16,384 × 1 | 0.73 GiB | 0.77 GiB | **0.89 GiB** | 1.68 GiB |
+| 131,072 × 1 | 5.83 GiB | 6.18 GiB | **7.12 GiB** | 13.41 GiB |
+| 1,048,576 × 1 | 46.62 GiB | 49.41 GiB | **56.98 GiB** | 107.25 GiB |
+
+These are planning estimates, not admission truth. The runtime must report measured allocated/resident MLA and indexer cache by Shard. Alpha configures a 16,384-token window, Q8_0 KV, and one session. Longer contexts and lower-bit KV are separate quality/resource certification gates.
+
+### 2.3 Activation seams and network
+
+A BF16 hidden-state boundary is 6,144 elements = 12,288 bytes/token before framing.
+
+- A 16,384-token prefill sends about 192 MiB per seam.
+- One decode token sends about 12 KiB per seam.
+- A 512-token decode sends about 6 MiB per seam.
+- Four nodes imply three serial seams.
+
+If a Shard boundary splits an IndexShare producer/consumer group, a 2,048-entry int32 top-k sideband can add up to 8 KiB/query before framing. The route planner should prefer boundaries that preserve complete IndexShare ownership groups. The protocol must still support and validate the named sideband because memory fit may force an internal group split.
+
+Decode bandwidth is small, but every generated token crosses all seams serially, so node count and per-hop latency dominate. Alpha requires a same-switch wired route: **2.5 GbE minimum and 10 GbE recommended**, with measured one-way/RTT, serialization, and queue latency. A 1 GbE route may be retained as fit-only evidence but is not the recommended alpha topology. Alpha records per-seam bytes, p50/p95 transfer latency, retries, and checksum failures; no speed claim is inferred from link rate alone.
+
+### 2.4 Storage
+
+The shortest alpha path allows every node to hold the complete six-file source GGUF while mapping/allocating only owned tensors. This minimizes artifact-transformation risk but costs 216.7 GB disk per node.
+
+Deterministic source-bound layer packages are a follow-up optimization. If needed before target fit, every package must retain:
+
+- source repository/revision and source file hashes;
+- exact owned tensor names, layer range, and endpoint role;
+- tokenizer/config identity;
+- deterministic package hash; and
+- proof that composing all packages matches the source tensor inventory.
+
+## 3. Current state and critical gaps
+
+### Completed foundation
+
+- DGR-001: immutable CPU contract plus separate signed ROCm diagnostic. CPU v1 remains `stop`; the GPU diagnostic establishes a viable fit/performance investigation lane but does not rewrite CPU evidence.
+- DGR-002: versioned backend-neutral gRPC/Protobuf Shard protocol with bounded fragments and compatibility checks.
+- Existing Meshnet: Tracker, contiguous Shards, Route Sessions/epochs, relay/direct transport, local Hot KV semantics in the reference backend, cancellation, telemetry, billing, and model-agnostic admission.
+
+### Missing before target alpha
+
+1. Exact GLM target/artifact manifest and memory-fit planner.
+2. A current llama.cpp pin proven to load and generate with the exact `UD-IQ1_S` artifact.
+3. A narrow decision on native GLM-5.2 DSA/IndexShare support. As observed 2026-07-13, merged llama.cpp PR #24770 loads GLM-5.2 through a dense-MLA compatibility path, while full IndexShare/DSA PR #25407 remains open and its generic sparse path can be slower than dense fallback. Generic CPU lightning-indexer support is merged; backend coverage remains uneven.
+4. A decode protocol amendment. `ActivationChunk` carries `TensorBundle`, but the current `DecodeStep` fast path carries only one `NamedTensor`; it cannot transport a hidden state plus GLM top-k sideband. Tail token/logit and sampling behavior also needs an explicit typed result contract.
+5. Correct range-owned GGUF loading and memory proof.
+6. GLM-specific boundary/KV/IndexShare semantics.
+7. Standalone native worker and Meshnet integration.
+8. Real target hardware route with no node individually able to admit the whole model.
+9. Locked target parity, usefulness, speed, failure, and cleanup evidence.
+
+### Donor policy
+
+`Mesh-LLM/mesh-llm` is a high-value test and patch donor. Its live GLM branch was observed with 261 llama.cpp patches, 167 named for GLM/DSA/MTP-related work. That is evidence of the problem's depth, not an acceptable maintained fork boundary.
+
+Audit and selectively reproduce the smallest independently understood pieces for:
+
+- GLM DSA graph semantics;
+- lightning indexer and sparse-attention tests;
+- IndexShare metadata/Full/Shared validation;
+- top-k sideband shape and lifecycle;
+- stage-local KV filtering; and
+- target parity/performance fixtures.
+
+Do not import Mesh-LLM routing, discovery, scheduler, public mesh, package manager, or full patch stack. Keep Meshnet as the sole control plane and collaborate narrowly upstream with llama.cpp/Mesh-LLM maintainers where practical.
+
+## 4. Revised roadmap
+
+## Phase 0 — lock the target before implementation
+
+### DGR-017: lock GLM-5.2 Max target and alpha contract
+
+Deliver:
+
+- machine-readable target manifest for official and GGUF revisions;
+- exact `UD-IQ1_S` file/size/hash inventory;
+- architecture/config/chat-template snapshot;
+- memory/KV/network planner with unified-memory de-duplication;
+- immutable alpha acceptance thresholds from section 5; and
+- current upstream/donor status report.
+
+Exit: target identity and alpha requirements are reviewable without downloading the model.
+
+## Phase 1 — establish a correct whole-model oracle
+
+### DGR-003: exact runtime recipe identity
+
+Extend the existing generic identity with GLM fields: DSA/IndexShare metadata, adapter version, reasoning template revision, activation bundle schema, KV dtype/layout, llama.cpp pin/patch hash, and target artifact manifest hash.
+
+### DGR-004: reproducible llama.cpp pin and narrow patch boundary
+
+Select a current exact upstream commit only after testing its stock GLM behavior. Add clean fetch/apply/build checks. Record every donor patch and whether it is adopted, rewritten, rejected, or waiting upstream.
+
+### DGR-018: certify whole-model GLM-5.2 runtime semantics
+
+On a 256-GiB-class reference host with at least 224 GiB runtime-accessible memory after OS reservation, or a measured equivalent:
+
+1. verify all six `UD-IQ1_S` shards;
+2. load with a stock pinned runtime and capture tensor/metadata warnings;
+3. prove whether DSA, IndexShare, shared expert, and Max template are actually active;
+4. add the minimum correctness patches/tests required;
+5. run deterministic prefill/decode and fixed Max-mode sentinel prompts; and
+6. sign the oracle recipe, output, telemetry, and limitations.
+
+Exit: one whole-model oracle exists for the same artifact/runtime semantics the distributed path will implement. “It emits text” is insufficient.
+
+## Phase 2 — build the generic local seam using small fixtures
+
+### DGR-005: range-owned GGUF tensors
+
+Keep dense Llama as a cheap structural fixture. Implement authoritative owned-tensor registration/loading, head/tail ownership, and measured resident-memory scaling. Design tensor classification so GLM adds explicit rules rather than unchecked name substitution.
+
+### DGR-006: architecture-defined boundary
+
+Implement named boundary bundles, F32 correctness lane, bounded fragmentation, and optional sidebands. Amend the decode fast path so it carries a versioned `TensorBundle` rather than one `NamedTensor`, while preserving a small one-tensor encoding. Define an explicit typed tail result for logits/token output and bind sampling/chat-template parameters to the recipe/request. Regenerate Python/C++ schema code and compatibility goldens. Dense fixture parity proves the seam mechanism, not GLM certification.
+
+Exit: two local processes can execute a small dense model with correct range ownership and boundary parity.
+
+## Phase 3 — add GLM-5.2 as the product adapter
+
+### DGR-019: implement and certify GLM-5.2 range/DSA/IndexShare semantics
+
+Deliver explicit support for:
+
+- 78 main layers and endpoint tensor ownership;
+- 256-expert MoE routing/top-8 and shared expert;
+- compressed MLA KV by owned layers;
+- DSA lightning indexer and sparse attention;
+- IndexShare metadata, Full producer, Shared consumer, and sideband behavior;
+- NextN/MTP tensor policy with MTP disabled or enabled explicitly;
+- shard-boundary planner aware of IndexShare ownership groups; and
+- whole-model versus two-stage parity against DGR-018.
+
+Exit: a same-host two-stage target run matches the locked oracle tolerance with real GLM computation in both stages. If the full target cannot fit on one host for this check, use a layer-reduced GLM architecture fixture for graph parity and defer full-artifact output parity to DGR-020; label the distinction explicitly.
+
+## Phase 4 — worker, KV, and Meshnet route
+
+Execute existing stories with GLM requirements included:
+
+1. DGR-007 — isolated local Hot KV keyed by `(Route Session, epoch)`, including DSA/IndexShare state.
+2. DGR-008 — standalone C++ gRPC worker.
+3. DGR-009 — Meshnet backend, capability, relay/direct, cancellation, and telemetry integration.
+4. DGR-010 — small-model local two-process acceptance.
+5. DGR-011 — real two-physical-machine route and heterogeneous fail-closed behavior.
+6. DGR-013 subset required by alpha — node loss, cancellation, stale epoch, restart, and memory/KV cleanup.
+
+Continuous batching (DGR-012) is deliberately not an alpha dependency. The first target release supports one admitted GLM route session; concurrency follows after target correctness and fit.
+
+## Phase 5 — target alpha gate
+
+### DGR-020: pass real distributed GLM-5.2 Max alpha acceptance
+
+Use at least two physical machines and enough aggregate usable memory to meet the locked target planner. No participating node may individually admit the complete target. All stages must report real compute and exact tensor ownership.
+
+Run the complete acceptance matrix in section 5, preserve raw logs/metrics/output, sign the evidence, and publish an explicit `alpha` or `stop` verdict. Thresholds cannot be weakened after results are known.
+
+## Phase 6 — post-alpha hardening
+
+After DGR-020 passes:
+
+1. DGR-012 — 1/2/4-session continuous batching and bounded admission.
+2. DGR-014 — final distributed GGUF versus reference-route performance decision.
+3. 32K, 128K, 200K, then 1M context certification with quantized KV.
+4. MTP/speculative decoding.
+5. Deterministic range packages to remove full-artifact replication.
+6. Additional backend compatibility classes and route topologies.
+7. DGR-016 — narrow upstream collaboration package, split by independently reviewable llama.cpp changes.
+8. DGR-015 — Qwen3/Qwen3-MoE only as later architecture expansion, not as the GLM alpha target.
+
+## 5. Locked alpha acceptance matrix
+
+These thresholds are set before target execution.
+
+### 5.1 Identity and fit
+
+- Exact official and GGUF repository revisions match the target manifest.
+- All six source GGUF sizes and LFS SHA-256 values verify.
+- Every route node reports owned tensor names/bytes, layer range, endpoint role, backend, KV recipe, and patch fingerprint.
+- Union of owned tensors equals the certified runtime-required tensor inventory; unintended overlap is zero.
+- No node's weight-plus-KV placement budget can hold the complete recipe.
+- Every node reserves at least `max(20% of physically usable memory, 8 GiB)` outside weight-plus-KV placement; measured peak scratch must remain inside that reserve.
+- Aggregate peak RSS/VRAM stays within physical budgets with no swap, overcommit, mmap-only, or double-counted unified-memory success claim.
+- Arithmetic-minimum topologies require exact contiguous tensor placement evidence; recommended alpha topology is 5×64 GiB or 3×96/128 GiB.
+- Unified RAM/VRAM is not double-counted.
+
+### 5.2 Semantic correctness
+
+- Logs and graph tests prove GLM MoE/shared-expert, DSA lightning indexer, sparse attention, and IndexShare Full/Shared paths are active; dense-attention compatibility fallback cannot satisfy alpha.
+- `reasoning_effort=max` is observable in the rendered template/API recipe.
+- F32 same-backend seam fixture: 32 greedy decode tokens exactly match the whole-model oracle and activation tolerance is locked by DGR-006.
+- Production seam on the fixed prompt corpus: greedy token agreement is at least 0.90 and mean compared-state/logit cosine similarity is at least 0.999 versus DGR-018, with no malformed or non-finite tensors.
+- Incompatible artifact, tokenizer, adapter, DSA metadata, boundary, activation, KV, backend class, or runtime patch fingerprints fail closed.
+
+### 5.3 End-to-end target run
+
+- Context configured to 16,384 tokens with Q8_0 MLA/indexer KV.
+- Fixed 4,096-token prompt lane completes prefill.
+- Route uses a same-switch wired network; 2.5 GbE is the alpha minimum and 10 GbE is recommended.
+- One Max-mode request generates at least 512 output tokens or reaches a valid natural EOS after at least 128 tokens.
+- Fixed coding, structured tool-call/JSON, and multi-step reasoning sentinels produce parseable, relevant outputs; raw prompts and outputs are retained for review.
+- OpenAI-compatible response includes stable model ID, finish reason, and token usage.
+
+### 5.4 Minimum useful performance
+
+On the declared minimum alpha topology after one warm-up:
+
+- median decode throughput is at least 0.5 generated token/s for the fixed Max-mode lane;
+- 4,096-token-prompt TTFT is at most 10 minutes;
+- no unexplained stall exceeds 60 seconds without progress telemetry;
+- per-stage compute, queue, KV, seam bytes/latency, RSS/VRAM, and backend timing are present; and
+- results are labeled by hardware/topology and are not generalized to other consumer systems.
+
+If output quality passes but the speed floor fails, verdict is `stop` for alpha and the evidence selects the next optimization target. It is not relabeled as success merely because the model loaded.
+
+### 5.5 Reliability and security
+
+- Two consecutive cold starts load, generate, release, and exit cleanly.
+- Cancellation during prefill and decode releases every stage's queued buffers and KV lease.
+- One worker loss aborts the route; alpha retries only from token zero on a new compatible route.
+- Stale epochs and duplicate step IDs are rejected.
+- Artifact paths stay outside `/home`; logs contain no secrets or unrestricted prompt payloads.
+- Synthetic workers and layer-reduced fixtures are labeled unit/integration coverage and cannot satisfy target alpha.
+
+## 6. First execution order
+
+The next unattended work should run in this order:
+
+1. DGR-017 — target contract, manifest, planner, and upstream status.
+2. DGR-003 — exact recipe identity.
+3. DGR-004 — current llama.cpp pin and minimal patch harness.
+4. Run in parallel:
+   - DGR-018 — whole-model oracle on a 256-GiB-class host with at least 224 GiB runtime-accessible memory.
+   - DGR-005 and DGR-006 — generic range/boundary seam on local small fixtures.
+5. DGR-019 — GLM semantics and parity after both parallel lanes pass.
+6. DGR-007 through DGR-011 — native worker and real transport route.
+7. Required DGR-013 failure subset.
+8. DGR-020 — real target alpha verdict.
+
+The first external hardware blocker is DGR-018, but DGR-005/DGR-006 proceed locally while that host is sourced. Do not download the full model until DGR-017's exact manifest and storage preflight pass.
+
+## 7. Sources checked on 2026-07-13
+
+Authoritative or primary:
+
+- Official model card and config: <https://huggingface.co/zai-org/GLM-5.2>
+- Official release/architecture blog: <https://z.ai/blog/glm-5.2>
+- Official code/documentation repository: <https://github.com/zai-org/GLM-5>
+- Official source revision API: <https://huggingface.co/api/models/zai-org/GLM-5.2>
+- Official GLM-5 technical report: <https://arxiv.org/abs/2602.15763>
+- Unsloth GGUF repository: <https://huggingface.co/unsloth/GLM-5.2-GGUF>
+- Unsloth local-run/quantization guide: <https://unsloth.ai/docs/models/glm-5.2>
+- llama.cpp GLM-5.2 support issue: <https://github.com/ggml-org/llama.cpp/issues/24730>
+- llama.cpp merged dense-MLA compatibility loader: <https://github.com/ggml-org/llama.cpp/pull/24770>
+- llama.cpp open GLM-5.2 DSA/IndexShare implementation: <https://github.com/ggml-org/llama.cpp/pull/25407>
+- llama.cpp merged generic CPU lightning indexer: <https://github.com/ggml-org/llama.cpp/pull/24231>
+- llama.cpp 1M-context discussion: <https://github.com/ggml-org/llama.cpp/discussions/24622>
+- IndexCache/IndexShare paper: <https://arxiv.org/abs/2603.12201>
+
+Donor/current implementation evidence:
+
+- Mesh-LLM repository: <https://github.com/Mesh-LLM/mesh-llm>
+- Mesh-LLM GLM branch noted by llama.cpp collaborator in issue #24730: `feat/jianyang-glm-52`
+
+Web/repository observations are pinned by date and must be refreshed in DGR-017 before implementation because upstream support is moving quickly.

.scratch/distributed-gguf-runtime/PRD.md

@@ -1,83 +1,382 @@
-# PRD: Distributed GGUF Runtime
+# PRD: Performant Concurrent Distributed GGUF Runtime
 
-## Summary
+## Overview
 
-Build a distributed inference runtime that can serve large, quality-first open models by combining torrent-style model artifact distribution with sticky multi-node Inference Routes and per-shard local Hot KV State.
+Build one lean native GGUF execution path that lets an Inference Route combine consumer machines to serve models larger than any one node can hold. Reuse the existing Meshnet control plane and llama.cpp/GGML execution engine. Adopt gRPC/HTTP2 and Protocol Buffers for the native Shard worker data plane rather than inventing a transport.
 
-The first runtime proof uses the existing PyTorch route because it exposes model internals and cache semantics more directly. GGUF/llama.cpp becomes the performance path after the route-session contract is proven.
+The program is benchmark-gated. GGUF is not assumed faster merely because it is quantized or uses a different file format. The first story compares the current Transformers/safetensors backend against whole-model llama.cpp on controlled model/hardware/quality lanes and locks a performance contract. Native distributed work proceeds only when GGUF provides a meaningful speed or fit benefit.
+
+The alpha target is now exact: `zai-org/GLM-5.2` at a pinned revision, the lowest published Unsloth `UD-IQ1_S` GGUF, and `reasoning_effort=max`, distributed across physical consumer machines. See [GLM-5.2-MAX-ALPHA-ROADMAP.md](GLM-5.2-MAX-ALPHA-ROADMAP.md). Dense Llama is a cheap structural fixture; Qwen is post-alpha expansion.
 
 ## Goals
 
-- Eliminate full-prompt recompute in distributed decode.
-- Keep decode activation seams proportional to `hidden_size`, not `context_length * hidden_size`.
-- Keep Hot KV State local to the node serving the relevant Shard.
-- Stream token deltas when feasible and always expose Generation Telemetry.
-- Add a local full-model GGUF backend for immediate CPU performance wins.
-- Define Model Artifact manifests so nodes can verify, seed, and advertise artifacts without depending on Hugging Face at request time.
-- Prototype an upstreamable llama.cpp/libllama layer-boundary API.
-- Use DeepSeek-V4-Flash as the first serious large-model target after smaller protocol smoke tests.
+- Execute one GGUF model across independently addressable contiguous Shards.
+- Retain Hot KV State locally for each Shard and isolate concurrent Route Sessions.
+- Batch compatible decode steps across active sessions for aggregate throughput.
+- Use consumer CPU, AMD, NVIDIA, Vulkan, Metal, and mixed routes only where a real certified forward passes.
+- Beat the current distributed safetensors route under a controlled performance contract or enable a larger otherwise-unroutable model at useful measured speed.
+- Keep the critical path to Meshnet plus a small pinned llama.cpp fork and standalone C++ worker.
+- Produce narrow upstream collaboration material for llama.cpp without placing Meshnet networking or economics inside upstream.
+- Pass an immutable GLM-5.2 Max `alpha`/`stop` gate with native MoE, DSA, IndexShare, parity, usefulness, speed, failure, and cleanup evidence.
+
+## Quality Gates
+
+Every story must:
+
+- Run its targeted `pytest` tests.
+- Run `python -m compileall packages tests` for Python changes.
+- Run `git diff --check`.
+- Keep default tests deterministic, model-download-free, API-credit-free, and GPU-free.
+- Preserve existing Transformers/safetensors behavior unless the story explicitly changes a versioned compatibility contract.
+
+Stories touching the native worker must also:
+
+- Build the pinned C++ target with CMake.
+- Run focused C++/protocol tests through CTest or the documented equivalent.
+- Verify the llama.cpp patch stack applies cleanly to the exact pinned commit.
+
+Real-model/hardware stories must:
+
+- Require `MESHNET_ENABLE_REAL_INFERENCE_TESTS=1`.
+- Use the machine-specific mounted-drive model path and the certified runtime environment; never place model artifacts under `/home`.
+- Record exact model revision, artifact hash, runtime recipe, hardware, driver/backend, commands, raw JSON metrics, and output-quality result.
+- Label synthetic tests as unit coverage rather than distributed acceptance.
+
+Before a story is marked complete, run the full deterministic `pytest -q` suite or record the exact pre-existing unrelated failure with a clean-tree reproduction.
+
+## Dependency Graph and Status
+
+Status as of 2026-07-14 (MAINT-003). Authoritative per-story status is
+`passes` in [prd.json](prd.json); closed issues live in
+`docs/issues/distributed-gguf-runtime/`, open and blocked issues in
+[issues/](issues/).
+
+```mermaid
+graph TD
+    classDef done fill:#c8e6c9,stroke:#2e7d32;
+    classDef blocked fill:#ffcdd2,stroke:#c62828;
+
+    DGR001[DGR-001 perf contract]:::done
+    DGR002[DGR-002 gRPC Shard protocol]:::done
+    DGR003[DGR-003 artifact/recipe identity]:::done
+    DGR004[DGR-004 pinned llama.cpp patch stack]:::done
+    DGR005[DGR-005 dense-Llama range ownership]:::done
+    DGR006[DGR-006 boundary input/output]:::done
+    DGR017[DGR-017 GLM-5.2 target/alpha contract]:::done
+    DGR018[DGR-018 whole-model GLM oracle]:::blocked
+    DGR019[DGR-019 GLM range/DSA/IndexShare]:::blocked
+    DGR020[DGR-020 distributed GLM alpha]:::blocked
+    DGR007[DGR-007 Hot KV State]
+    DGR008[DGR-008 C++ gRPC worker]
+    DGR009[DGR-009 Meshnet integration]
+    DGR010[DGR-010 local two-process acceptance]
+    DGR011[DGR-011 two-machine route]
+    DGR012[DGR-012 continuous batching]
+    DGR013[DGR-013 failure/cancel/restart]
+    DGR014[DGR-014 release gate]
+    DGR015[DGR-015 Qwen3 adapter]
+    DGR016[DGR-016 upstream package]
+
+    DGR002 --> DGR003
+    DGR017 --> DGR003
+    DGR001 --> DGR004
+    DGR017 --> DGR004
+    DGR003 --> DGR005
+    DGR004 --> DGR005
+    DGR002 --> DGR006
+    DGR005 --> DGR006
+    DGR001 --> DGR017
+    DGR002 --> DGR017
+    DGR003 --> DGR018
+    DGR004 --> DGR018
+    DGR017 --> DGR018
+    DGR005 --> DGR019
+    DGR006 --> DGR019
+    DGR018 --> DGR019
+    DGR006 --> DGR007
+    DGR019 --> DGR007
+    DGR002 --> DGR008
+    DGR003 --> DGR008
+    DGR004 --> DGR008
+    DGR006 --> DGR008
+    DGR007 --> DGR008
+    DGR003 --> DGR009
+    DGR008 --> DGR009
+    DGR009 --> DGR010
+    DGR010 --> DGR011
+    DGR007 --> DGR012
+    DGR009 --> DGR012
+    DGR010 --> DGR012
+    DGR008 --> DGR013
+    DGR009 --> DGR013
+    DGR001 --> DGR014
+    DGR011 --> DGR014
+    DGR012 --> DGR014
+    DGR013 --> DGR014
+    DGR014 --> DGR015
+    DGR010 --> DGR016
+    DGR007 --> DGR020
+    DGR008 --> DGR020
+    DGR009 --> DGR020
+    DGR011 --> DGR020
+    DGR013 --> DGR020
+    DGR017 --> DGR020
+    DGR018 --> DGR020
+    DGR019 --> DGR020
+```
+
+- **Done (`passes: true`):** DGR-001, DGR-002, DGR-003, DGR-004, DGR-005,
+  DGR-006, DGR-017.
+- **Blocked on hardware:** DGR-018 requires a 256-GiB-class host with at least
+  224 GiB runtime-accessible memory and 250 GB free storage outside `/home`;
+  no such host is currently available (development host: 124.9 GiB MemTotal).
+  Exact preflight output: [evidence/DGR-018/BLOCKED.md](evidence/DGR-018/BLOCKED.md).
+  DGR-019 (needs the DGR-018 oracle) and DGR-020 (needs DGR-018/DGR-019 plus
+  multiple physical consumer nodes) are blocked transitively.
+- **Consequence of the graph as written:** DGR-007 depends on DGR-019, so every
+  remaining story (DGR-007 through DGR-016) is transitively blocked on the
+  256-GiB host. Unblocking the generic dense pipeline without that host would
+  require an explicit re-planning decision to relax the DGR-007 → DGR-019
+  dependency; that decision is out of scope for maintenance and has not been
+  made.
+
+## User Stories
+
+### DGR-001: Lock the safetensors-versus-GGUF performance contract
+**Description:** As a runtime engineer, I need a controlled baseline so that GGUF work proceeds from measured speed, memory, and quality rather than reputation.
+
+**Acceptance Criteria:**
+- [ ] Benchmark the same model architecture/revision, machine, prompts, context lengths, output lengths, sampling policy, and concurrency across the current Transformers/safetensors recipe and whole-model llama.cpp recipes.
+- [ ] Separate correctness/quality lanes from quantized performance/fit lanes instead of claiming BF16 and Q4 are numerically equivalent.
+- [ ] Report TTFT, prefill tok/s, decode tok/s, p50/p95 latency, aggregate throughput, RSS, VRAM, artifact size, failures, and output drift in machine-readable JSON.
+- [ ] Add concurrency levels 1 and 4 where memory permits.
+- [ ] Write a versioned performance contract consumed by later release gates, including an explicit stop condition when llama.cpp/GGUF has no meaningful speed or fit benefit.
+
+### DGR-002: Adopt the versioned gRPC Shard protocol
+**Description:** As a node developer, I need a battle-proven streaming protocol so that Python and C++ Shards communicate without a custom socket protocol.
+
+**Acceptance Criteria:**
+- [ ] Add a Protocol Buffers schema for capability, health, session stream, release, and cancellation operations.
+- [ ] Define one long-lived bidirectional gRPC stream per Route Session Activation Seam with deadlines, cancellation, flow control, and structured errors.
+- [ ] Define bounded chunking for prefill and a small decode fast path.
+- [ ] Carry schema version, request/work ID, Route Session ID, route epoch, artifact/recipe fingerprint, Shard range/effective start, phase, position, idempotency step, cache expectation, compression, and checksum.
+- [ ] Define a versioned named-tensor bundle with per-tensor name, shape, dtype, byte order, and payload fragments.
+- [ ] Add generated-schema round-trip and compatibility tests in Python and C++.
+
+### DGR-003: Define exact Artifact and runtime recipe identity
+**Description:** As the Tracker, I need exact compatibility identity so that only numerically and operationally compatible Shards form an Inference Route.
+
+**Acceptance Criteria:**
+- [ ] Separate weight quantization, activation dtype, compute dtype, KV dtype/layout, tokenizer revision, architecture adapter, backend, and runtime version.
+- [ ] Bind derivative or split artifacts to an exact source Model Artifact hash and Shard range.
+- [ ] Produce a stable compatibility fingerprint used by capability admission and the gRPC handshake.
+- [ ] Fail closed on mismatched artifact, tokenizer, architecture, range, boundary schema, activation recipe, or cache layout.
+- [ ] Keep unsupported recipes registered-but-dark until a real distributed forward certifies them.
+
+### DGR-004: Create the reproducible pinned llama.cpp patch stack
+**Description:** As a maintainer, I need a small auditable fork boundary so that upstream updates do not turn the runtime into an unmaintainable stitched codebase.
+
+**Acceptance Criteria:**
+- [ ] Pin one exact llama.cpp commit through a reproducible source dependency mechanism.
+- [ ] Store a numbered minimal patch stack separately from Meshnet networking code.
+- [ ] Add a build script that applies/checks patches and builds the standalone worker without manual source copying.
+- [ ] Record upstream file/ABI assumptions and fail clearly when the pin changes.
+- [ ] Preserve upstream license and attribution notices.
+- [ ] Add a clean rebuild smoke test that does not download a model.
+
+### DGR-005: Implement dense-Llama range-aware GGUF ownership
+**Description:** As a node, I need to map only my assigned dense-Llama Shard so that aggregate consumer memory can hold a model larger than one node.
+
+**Acceptance Criteria:**
+- [ ] Register and allocate only `blk.N.*` tensors in the assigned range.
+- [ ] Load embeddings only for the head and final norm/LM head only for the tail, including tied embeddings.
+- [ ] Prefer range-aware mapping from one exact source GGUF; if derivative sub-GGUFs are used temporarily, verify source/slice hashes and avoid claiming final artifact semantics.
+- [ ] Report authoritative loaded range and endpoint ownership from the model, not operator CLI claims.
+- [ ] Demonstrate mapped/resident memory scales with owned tensors rather than full model size.
+
+### DGR-006: Implement architecture-defined boundary input/output
+**Description:** As a Shard, I need to consume and emit the correct transformer boundary state so that disjoint processes reproduce whole-model execution.
+
+**Acceptance Criteria:**
+- [ ] Head accepts token IDs and owns token embedding.
+- [ ] Middle/tail bypass token embedding and accept the named boundary bundle.
+- [ ] Amend the decode fast path from one `NamedTensor` to a versioned `TensorBundle`, preserving compact one-tensor compatibility and regenerating Python/C++ protocol goldens.
+- [ ] Define a typed tail logits/token result with sampling and chat-template/reasoning identity.
+- [ ] Non-tail emits the unnormalized architecture-defined residual/boundary before final norm/head and before tail-only row pruning.
+- [ ] Tail emits logits or token output through an explicit sampling contract.
+- [ ] Dense-Llama whole-model versus two-range prefill and greedy-decode parity passes the documented tolerance.
+- [ ] The adapter interface fails closed for uncertified architectures.
+
+### DGR-007: Add isolated concurrent local Hot KV State
+**Description:** As a client, I need concurrent Route Sessions to retain independent per-Shard cache so that one request cannot clear or corrupt another.
+
+**Acceptance Criteria:**
+- [ ] Map `(Route Session ID, route epoch)` to an isolated llama sequence or bounded context.
+- [ ] Allocate KV only for owned layers.
+- [ ] Support prefill append, decode append, truncate, release, TTL/LRU eviction, and explicit cache-miss response.
+- [ ] Reject stale epochs and incompatible cache recipes.
+- [ ] At least four concurrent sessions on a small model complete without token or KV cross-talk.
+- [ ] Cancellation/release of one session leaves other sessions intact and memory returns to the configured budget.
+
+### DGR-008: Build the standalone C++ gRPC Shard worker
+**Description:** As a node runtime, I need one supervised native process so that llama.cpp internals remain behind a stable project-owned protocol.
+
+**Acceptance Criteria:**
+- [ ] Worker exposes capability, health, session stream, release, cancellation, and metrics services from DGR-002.
+- [ ] Worker loads one exact Artifact/recipe/Shard identity and refuses mismatched requests.
+- [ ] Streaming path enforces bounded messages, flow control, deadlines, idempotency, and independent session cancellation.
+- [ ] Worker does not expose raw llama.cpp RPC or arbitrary GGML graph execution.
+- [ ] Graceful shutdown releases sessions; crash behavior is bounded and observable.
+- [ ] Python integration tests run against a fake model mode without model downloads.
+
+### DGR-009: Integrate the native worker with Meshnet
+**Description:** As the existing node service, I need a GGUF Shard backend adapter so that the Tracker, relay, billing, telemetry, and capability admission remain the sole control plane.
+
+**Acceptance Criteria:**
+- [ ] Implement the existing model-backend surface without changing Transformers behavior.
+- [ ] Registration carries exact validated GGUF recipe, Shard, backend and concurrency/KV capacity.
+- [ ] Tracker forms only complete compatible routes and keeps uncertified recipes dark.
+- [ ] Direct routes use gRPC streams; relayed routes carry the same versioned protobuf frames as opaque binary through the existing relay seam.
+- [ ] Existing request/work IDs, cancellation, Generation Telemetry, billing, and per-node attribution remain correlated.
+- [ ] No vLLM, Nakshatra, prima.cpp, or custom-engine control plane becomes a core dependency.
+
+### DGR-010: Pass local real-model two-process acceptance
+**Description:** As a release engineer, I need real local distributed parity before involving network variability.
+
+**Acceptance Criteria:**
+- [ ] Two local worker processes open disjoint dense-Llama ranges from the certified Artifact.
+- [ ] Prefill and at least 32 greedy decode tokens match whole-model llama.cpp within the certified tolerance.
+- [ ] Each worker retains only its own tensors and Hot KV State.
+- [ ] Four concurrent Route Sessions pass isolation and cleanup checks.
+- [ ] Report TTFT, prefill/decode throughput, seam bytes/latency, worker RSS/VRAM, KV memory, batch size, and queue time.
+- [ ] Killing one worker produces a bounded structured failure rather than a deadlock.
+
+### DGR-011: Pass a real heterogeneous two-machine route
+**Description:** As a consumer-hardware operator, I need two physical machines to execute one GGUF model so that the distributed claim is real.
+
+**Acceptance Criteria:**
+- [ ] Tracker selects two physical nodes with disjoint Shards and one exact certified recipe/compatibility class.
+- [ ] Actual CPU/GPU execution occurs on both nodes; synthetic workers do not satisfy acceptance.
+- [ ] Prefill/decode, concurrent-session isolation, telemetry, cancellation, and cleanup pass over the real transport/relay path.
+- [ ] Exact hardware, network, backend, model hash, route, commands, and raw metrics are recorded.
+- [ ] A model or recipe larger than one participating node's admitted memory is exercised when available.
+- [ ] Output drift is measured and incompatible mixed backends fail closed.
+
+### DGR-012: Implement continuous batching and bounded admission
+**Description:** As a node operator, I need active sessions batched safely so that concurrency increases aggregate throughput rather than serializing every request.
+
+**Acceptance Criteria:**
+- [ ] Node scheduler admits sessions against weight, KV, scratch, and queue budgets.
+- [ ] Compatible decode steps from multiple sessions form llama.cpp batches while preserving per-session positions and outputs.
+- [ ] Prefill does not starve decode; scheduling policy and bounds are explicit.
+- [ ] Backpressure prevents unbounded queued activations or KV growth.
+- [ ] Capability telemetry reports active sessions, queue depth, batch occupancy, KV pressure, prefill/decode rates, and rejected admissions.
+- [ ] Concurrency 1/2/4/8 benchmark identifies saturation and shows no cross-session corruption.
+
+### DGR-013: Harden failure, cancellation, and restart semantics
+**Description:** As a client, I need failures to be bounded and explicit so that distributed speed does not come with hanging or corrupted generations.
+
+**Acceptance Criteria:**
+- [ ] Deadlines and heartbeat/health loss terminate blocked stream operations.
+- [ ] Cancellation propagates across every Shard and releases local KV and queued buffers.
+- [ ] Duplicate steps are idempotent; uncertain mutations are never replayed silently.
+- [ ] Alpha failover restarts from token zero on a newly compatible route rather than importing unverified KV.
+- [ ] Worker death, stream reset, malformed bundle, stale epoch, and cache miss tests pass.
+- [ ] Billing/work records distinguish completed, cancelled, failed, and unverified work.
+
+### DGR-014: Enforce the GGUF-versus-safetensors release gate
+**Description:** As the product owner, I need an end-to-end comparison so that the native runtime ships only if it advances model access or performance.
+
+**Acceptance Criteria:**
+- [ ] Run current distributed safetensors and distributed GGUF routes on the same certified model/hardware/network scenario where technically comparable.
+- [ ] Report quality, TTFT, prefill/decode throughput, aggregate concurrency throughput, p95 latency, seam cost, memory, KV pressure, failures, and cleanup.
+- [ ] Evaluate against the DGR-001 performance contract without changing thresholds after seeing results.
+- [ ] Ship recommendation is one of: promote GGUF, optimize a measured bottleneck with a new bounded task, or stop the native track.
+- [ ] Results clearly separate quantization gains from transport/runtime gains.
+
+### DGR-015: Add and certify a Qwen3/Qwen3-MoE adapter
+**Description:** As a client seeking top models, I need a separately certified MoE-capable architecture after the dense runtime proves stable.
+
+**Acceptance Criteria:**
+- [ ] Implement explicit tensor ownership, router/top-k, expert/shared-expert, Q/K normalization, boundary bundle, and cache semantics for the selected Qwen3 family recipe.
+- [ ] Do not reuse the dense-Llama adapter through unchecked name substitutions.
+- [ ] Whole-model versus distributed prefill/decode parity passes the architecture-specific tolerance.
+- [ ] Expert memory ownership and communication are measured.
+- [ ] Real consumer-hardware acceptance and capability admission pass before the recipe becomes routable.
+
+### DGR-016: Produce the upstream llama.cpp collaboration package
+**Description:** As a maintainer, I need narrow upstreamable proposals so that our patch burden can shrink without asking llama.cpp to own Meshnet networking.
+
+**Acceptance Criteria:**
+- [ ] Separate generic llama.cpp hooks from Meshnet protocol/control-plane code.
+- [ ] Prepare minimal reproducible examples and tests for range-aware loading, boundary input/output, and layer-filtered KV.
+- [ ] Compare the proposal with Nakshatra and prima.cpp evidence and explain why the API is generally useful.
+- [ ] Preserve one scoped commit/patch per concern against the exact upstream pin.
+- [ ] Produce an outreach document suitable for Georgi/llama.cpp maintainers; actual sending remains a human action.
+
+### DGR-017: Lock the GLM-5.2 Max target and alpha contract
+**Description:** Pin exact official/GGUF revisions, `UD-IQ1_S` files and hashes, Max-mode semantics, resource accounting, and immutable target thresholds before implementation results exist.
+
+**Acceptance Criteria:**
+- [ ] Produce machine-readable target, resource, upstream-status, and immutable acceptance contracts without downloading full weights.
+- [ ] Distinguish the 224-GiB hard-fit floor from the recommended 5×64 or 3×96/128 topology, using Q8 KV, 20%/8-GiB node reserves, and a wired 2.5-GbE minimum.
+- [ ] Count unified RAM/VRAM once and test revision/shard/byte/threshold mutation failures.
+
+### DGR-018: Certify whole-model GLM-5.2 runtime semantics
+**Description:** Establish the exact IQ1_S oracle on a 256-GiB-class host with at least 224 GiB runtime-accessible memory; lock Q8_0 MLA/indexer KV and native target semantics before distributed parity work.
+
+**Acceptance Criteria:**
+- [ ] Verify the complete artifact and prove native MoE/shared expert, DSA, IndexShare, KV, NextN policy, and Max-template behavior.
+- [ ] Dense/replicated compatibility fallback cannot become the oracle merely because it emits text.
+
+### DGR-019: Implement and certify GLM-5.2 range, DSA, and IndexShare semantics
+**Description:** Add explicit target-model tensor, graph, boundary, sideband, and local-KV ownership after the generic dense seam.
+
+**Acceptance Criteria:**
+- [ ] Preserve MoE/shared expert, DSA lightning indexer/sparse attention, and IndexShare Full/Shared semantics across contiguous Shards.
+- [ ] Pass locked fixture/target parity and measured per-Shard memory ownership; never claim full-target parity from a reduced fixture.
+
+### DGR-020: Pass real distributed GLM-5.2 Max alpha acceptance
+**Description:** Run the exact lowest-quant target through Meshnet on enough physical consumer nodes that no node can admit the whole recipe.
+
+**Acceptance Criteria:**
+- [ ] Pass the immutable identity, semantic, parity, Max-mode usefulness, 0.5 token/s, TTFT, reliability, and mounted-storage gates.
+- [ ] Preserve signed raw evidence and emit `alpha` only if every target criterion passes; otherwise emit `stop`.
+
+## Functional Requirements
+
+1. The public distributed primitive is an ordered Inference Route of contiguous Shards.
+2. The native runtime uses llama.cpp/GGML; vLLM remains optional as a complete managed provider.
+3. Native worker communication uses gRPC/HTTP2 and Protocol Buffers with one stable stream per Route Session Activation Seam.
+4. Artifact identity, runtime recipe, boundary schema, activation dtype and cache layout must match exactly before routing.
+5. Hot KV State remains local to the node serving the Shard.
+6. Multiple Route Sessions must execute concurrently without shared-cache corruption.
+7. Nodes batch compatible active decode steps and enforce bounded admission/backpressure.
+8. Unsupported architectures and hardware recipes remain non-routable until real certification passes.
+9. Default tests never download models or require GPUs; real tests are explicit and preserve artifacts off `/home`.
+10. The release decision is based on measured performance, fit, quality, concurrency, and reliability relative to the safetensors baseline.
 
 ## Non-Goals
 
-- No centralized hot KV cache in the per-token decode path.
-- No automatic route repair in alpha.
-- No permanent llama.cpp fork as the intended architecture.
-- No GLM-5.2 or Ornith first; they remain follow-up support audits.
-- No transport rewrite to QUIC/WebRTC before route/session semantics are proven.
+- Forking vLLM or importing its PagedAttention/Torch distributed runtime.
+- Adopting Nakshatra, prima.cpp, llama-gguf, LiGGUF, or GPUStack as the control plane.
+- Public WAN tensor/expert parallel collectives.
+- QUIC, WebRTC, or a custom socket protocol.
+- Automatic KV migration or mid-generation route repair in the first release.
+- Speculative decoding or disaggregated prefill before the core release gate.
+- Supporting every GGUF architecture before the exact GLM-5.2 target; Qwen3-family certification is post-alpha.
+- A marketing-scale model demo that bypasses parity, concurrency, admission, or performance gates.
 
-## Resolved Decisions
+## Success Metrics
 
-- Public-network Shards are contiguous transformer layer ranges.
-- Tensor/ring parallelism belongs inside one trusted node, one colocated pod, or a future composite node abstraction.
-- Hot KV State is local to route nodes; Prefix Snapshots are optional cold recovery/reuse artifacts.
-- PyTorch distributed KV/session semantics are proven before llama.cpp distributed execution.
-- Streaming responses are preferred; Generation Telemetry is mandatory.
-- llama.cpp/GGUF work targets upstreamable `libllama`/ggml hooks.
-- Alpha fails Route Sessions on route-node loss.
-- v1 activation transfer stays on binary HTTP.
+- A real model larger than one admitted node can execute across consumer machines when suitable hardware/artifacts are available.
+- Four or more concurrent sessions complete without cross-talk; hardware-specific saturation is measured.
+- Distributed GGUF passes the locked performance/fit contract against the existing safetensors route.
+- Worker and Tracker recover all resources after completion, cancellation, malformed input, and node failure.
+- The critical runtime remains Meshnet plus one standalone worker and a small auditable llama.cpp patch stack.
 
-## Target User Experience
+## Open Questions
 
-A client sends an OpenAI-compatible request. The Gateway or Tracker Node accepts the request, creates a Route Session, and streams token deltas when supported. The client receives live Generation Telemetry for route phase, prefill progress, generated token count, rolling tokens/sec, route health, and failure reason.
-
-If a route node drops in alpha, the request fails clearly. A retry starts a new Route Session from scratch.
-
-## Runtime Shape
-
-```text
-client request
-  -> Gateway / Tracker Node creates Route Session
-  -> Tracker selects sticky Inference Route
-  -> prefill:
-       prompt chunks move through Shards
-       each node appends local Hot KV State
-  -> decode:
-       one-step activation moves through Shards
-       each node reads/appends local Hot KV State
-       tail returns token/logits
-  -> client receives streamed token deltas where possible
-  -> Generation Telemetry continues until complete or failed
-```
-
-## Milestones
-
-| Milestone | Outcome | Issues |
-|---|---|---|
-| M1 — Session protocol proof | Stub route has stable Route Sessions, prefill/decode split, telemetry, and streaming contract | 01, 02, 03 |
-| M2 — PyTorch reference route | Distributed PyTorch decode uses local per-shard cache and stops full-prompt recompute | 04 |
-| M3 — Local GGUF performance path | Single-node GGUF backend serves through the node API and reports backend metadata | 05 |
-| M4 — Artifact plane | Model Artifact manifest supports verification, layer mapping, and node advertisement | 06 |
-| M5 — llama.cpp collaboration proof | Localhost layer-boundary prototype identifies upstreamable llama.cpp/libllama API | 07 |
-| M6 — Networked GGUF route | Multi-node GGUF route uses the resolved protocol and fails cleanly on node loss | 08 |
-| M7 — First large model | DeepSeek-V4-Flash support path is audited and converted into follow-up runtime tasks | 09 |
-
-## Acceptance Criteria
-
-- A two-node route can prefill once and decode without resending full prompt activations.
-- Decode seam payload is one token/hidden-state step after prefill.
-- Route Session telemetry is visible before first token and during decode.
-- Streaming token deltas work where the backend supports them.
-- Route-node loss produces a structured alpha failure and does not attempt unsafe repair.
-- A local GGUF model can serve via the node API.
-- A Model Artifact manifest can prove which Shards a node can serve.
-- DeepSeek-V4-Flash has a written support recommendation: PyTorch, vLLM/SGLang, llama.cpp/GGUF, or blocked.
+- DGR-001 remains immutable. DGR-017 locks the exact GLM-5.2/`UD-IQ1_S` target and alpha thresholds without rewriting DGR-001 evidence.
+- Final hardware-specific concurrency and useful-speed thresholds are locked by measured baselines rather than guessed globally.
+- Upstream llama.cpp acceptance is desirable but not a prerequisite for the first narrow pinned fork.

.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md

@@ -0,0 +1,319 @@
+# Ralph execution context: Performant Concurrent Distributed GGUF Runtime
+
+Status: authoritative context for every fresh Ralph iteration
+Last updated: 2026-07-13
+
+## Mandatory startup sequence
+
+Before changing code, every Ralph agent must:
+
+1. Read this file completely.
+2. Read the selected issue under `.scratch/distributed-gguf-runtime/issues/`.
+3. Read `.scratch/distributed-gguf-runtime/GLM-5.2-MAX-ALPHA-ROADMAP.md`, `.scratch/distributed-gguf-runtime/ADR-0020-distributed-gguf-runtime.md`, and the relevant part of `architecture.md`.
+4. Read `.claude/memory/MEMORY.md` and root `CONTEXT.md` for current project vocabulary and constraints.
+5. Inspect the current implementation and tests; do not assume historical scratch text describes live code.
+6. Read the evidence/handoff directories for every declared dependency.
+7. Inspect `git status` and preserve all pre-existing working-tree changes.
+
+A fresh Ralph iteration has no conversational memory. These files are the context contract.
+
+## Story sizing and interruption rule
+
+Each story is intended to fit one focused Ralph context. Before implementation, estimate whether every acceptance criterion can be completed and verified in the current iteration.
+
+If the story is too large, an external dependency is unavailable, or the context/provider limit prevents completion:
+
+- Do not weaken criteria.
+- Do not mark the issue done or set `passes: true`.
+- Avoid leaving an unverified cross-cutting partial implementation when a smaller safe spike is possible.
+- Write `evidence/<TASK-ID>/DECOMPOSITION.md` or `BLOCKED.md` with the exact blocker, current verified state, proposed child stories, dependency graph and rollback/continuation instructions.
+- Stop for supervised review.
+
+If interrupted after code changes, record every changed file, command result and unresolved invariant so the next fresh loop can verify rather than guess.
+
+## Product objective
+
+Build performant, concurrent distributed inference that combines consumer machines to serve top open models that exceed one node's RAM/VRAM.
+
+The alpha target is the exact pinned GLM-5.2 `UD-IQ1_S` artifact served with `reasoning_effort=max` across physical consumer machines. Dense Llama is a structural fixture. Synthetic workers, dense-attention compatibility fallback, a smaller model, or a single host cannot satisfy target alpha. The immutable target contract and resource envelope are in `GLM-5.2-MAX-ALPHA-ROADMAP.md`.
+
+A distributed demo is not success. The product must provide:
+
+- Useful measured prefill and decode speed.
+- Multiple concurrent Route Sessions.
+- No KV/token cross-talk.
+- Bounded memory, queues, cancellation and failures.
+- Real execution on every participating node.
+- A model-fit or performance advantage over the current Transformers/safetensors route.
+
+## Critical-path architecture
+
+```text
+Existing Meshnet control plane
+        |
+Versioned Protobuf over gRPC/HTTP2
+        |
+Project-owned standalone C++ Shard worker
+        |
+Small exact-commit llama.cpp patch stack
+```
+
+Meshnet remains the only control plane and owns:
+
+- Tracker registration, Coverage Map, route selection and route epochs.
+- Route Sessions and Activation Seams.
+- Direct/relay routing.
+- Capability admission.
+- Cancellation, Generation Telemetry and backpressure.
+- Billing, validation and per-node work attribution.
+
+Do not introduce another scheduler/control plane from vLLM, Nakshatra, prima.cpp, llama-gguf, GPUStack or another project.
+
+## Runtime decisions that are not open
+
+1. Public-network Shards are contiguous transformer layer ranges.
+2. llama.cpp/GGML is the native GGUF execution substrate.
+3. The project owns a small standalone worker and a narrow pinned llama.cpp patch stack.
+4. The native Shard protocol is Protocol Buffers over gRPC/HTTP2.
+5. One long-lived bidirectional stream serves one Route Session Activation Seam.
+6. The public activation boundary is a versioned named-tensor bundle.
+7. Hot KV State remains local to the node serving the Shard.
+8. `(Route Session ID, route epoch)` maps to an isolated llama sequence or bounded context.
+9. Concurrency uses continuous batching of compatible active sessions inside each node.
+10. Transformers/safetensors remains the correctness and performance baseline.
+11. vLLM may be an optional complete managed provider and concept donor; it is not forked into public Shards.
+12. Tensor/expert collectives are deferred to a trusted composite provider, not public WAN routes.
+13. Unsupported architectures/backends remain registered-but-dark until real certification passes.
+14. Alpha failure retries from token zero; unverified KV is never migrated silently.
+15. Model artifacts must remain on mounted-drive storage and never under `/home`.
+16. Unified system RAM and integrated-GPU memory are one physical pool and must never be double-counted for admission.
+17. Alpha requires native GLM-5.2 MoE, DSA, and IndexShare semantics; MTP/speculative decoding and 1M-context certification are post-alpha.
+18. DGR-006 amends the decode fast path to carry a versioned `TensorBundle` and defines a typed tail logits/token result; the current single-`NamedTensor` fast path is insufficient for GLM sidebands.
+19. Alpha reserves at least `max(20% of physical usable memory, 8 GiB)` per node outside weight-plus-Q8-KV placement and uses a same-switch wired 2.5 GbE minimum route.
+
+Changing one of these requires an explicit ADR update and human review, not an incidental story implementation.
+
+## Performance discipline
+
+GGUF performance is a hypothesis. Never write “GGUF is faster” without measurements.
+
+DGR-001 locks controlled benchmark lanes and thresholds. DGR-014 enforces the final distributed comparison.
+
+Always distinguish:
+
+- Weight quantization from activation/compute/KV dtype.
+- Runtime/kernel gains from quantization/model-fit gains.
+- Single-request latency from aggregate concurrency throughput.
+- Synthetic unit coverage from real distributed acceptance.
+
+Required metrics where applicable:
+
+```text
+TTFT
+prefill tokens/sec
+decode tokens/sec
+aggregate throughput
+p50/p95 latency
+seam bytes and latency
+queue and batch occupancy
+RSS and VRAM
+KV pressure
+output-quality drift
+failures and cleanup
+```
+
+Do not weaken or move performance thresholds after seeing implementation results.
+
+## Transport discipline
+
+Do not invent a raw TCP protocol, new WebSocket protocol, QUIC layer or bespoke binary control format.
+
+The `.proto` schema is the semantic contract. Direct transport uses gRPC. Existing relay infrastructure may carry the same serialized protobuf frames as opaque binary.
+
+Protocol requirements:
+
+- Schema/version negotiation.
+- Request/work ID.
+- Route Session ID and route epoch.
+- Exact Model Artifact/runtime recipe fingerprint.
+- Shard range and effective overlap-safe start.
+- Prefill/decode/release/cancel phases.
+- Position/token range and idempotency step.
+- Named tensors with shape, dtype, byte order and bounded fragments.
+- Compression/checksum.
+- Cache expectation/result.
+- Deadlines, cancellation, flow control and structured status.
+
+Avoid per-token channel creation and unbounded unary payloads. Generated code and build tooling must be reproducible; do not require manual copying.
+
+## Native runtime discipline
+
+Reuse llama.cpp for GGUF, mmap, kernels, architecture graphs, tokenizer, KV, sequences and heterogeneous backends.
+
+The project patch stack is limited to:
+
+- Range-aware tensor registration/loading.
+- Endpoint-specific embedding/final head ownership.
+- Architecture-defined intermediate input/output.
+- Intermediate output before final norm/head.
+- Layer-filtered KV and session mapping.
+
+Do not place Meshnet routing, transport, billing or authentication inside llama.cpp. Keep patches numbered, scoped, pinned and upstreamable.
+
+Dense Llama is first only as a cheap range/boundary fixture. GLM-5.2 is the explicit product adapter and alpha target immediately afterward. Qwen3/Qwen3-MoE is post-alpha. Do not generalize through unchecked tensor-name substitutions.
+
+## Existing code seams to inspect first
+
+- `packages/node/meshnet_node/model_backend.py` — backend abstraction.
+- `packages/node/meshnet_node/torch_server.py` — reference ranged execution and session behavior.
+- `packages/node/meshnet_node/activation_compression.py` — current activation framing/compression.
+- `packages/node/meshnet_node/route_session_benchmark.py` — existing benchmark infrastructure.
+- `packages/tracker/meshnet_tracker/server.py` — registration, route and proxy behavior.
+- `packages/tracker/meshnet_tracker/capability.py` — fail-closed capability admission.
+- `tests/test_real_model_backend.py` — real backend coverage.
+- `tests/test_tracker_routing.py` — route/session behavior.
+- `tests/test_tracker_capability_admission.py` — recipe admission.
+- `tests/test_route_session_benchmark.py` and `tests/test_manual_route_benchmark.py` — benchmark patterns.
+- `docs/adr/0008-binary-activation-wire-format.md` — existing wire compatibility.
+- `docs/adr/0012-start-layer-overlapping-shards.md` — effective start semantics.
+- `docs/adr/0022-sharded-per-node-kv-cache.md` — Hot KV State contract.
+- `docs/adr/0023-model-agnostic-node-capability-admission.md` — certification/admission.
+
+Do not edit generated `build/`, `__pycache__`, egg-info, Ralph logs or unrelated scratch features.
+
+## Planned source layout
+
+Use these paths unless current code inspection proves a better project-consistent location. If changed, document the reason in task evidence.
+
+```text
+packages/node/native/
+  proto/shard_runtime.proto
+  cmake/
+  llama/
+    UPSTREAM_COMMIT
+    patches/
+  gguf_worker/
+  tests/
+
+packages/node/meshnet_node/
+  native_protocol/
+  gguf_backend.py
+  runtime_recipe.py
+
+.scratch/distributed-gguf-runtime/evidence/<TASK-ID>/
+  README.md
+  commands.txt
+  results.json or other machine-readable evidence
+```
+
+Generated protobuf/C++ build outputs belong in build directories unless packaging explicitly requires checked-in generated Python modules. The story must document the generation command and version.
+
+## Story output map
+
+| Story | Required durable outputs |
+|---|---|
+| DGR-001 | benchmark harness/tests; `evidence/DGR-001/performance-contract.json`; raw/summary benchmark evidence |
+| DGR-002 | `packages/node/native/proto/shard_runtime.proto`; reproducible Python/C++ generation/build wiring; protocol round-trip/compatibility tests; `evidence/DGR-002/` |
+| DGR-003 | exact runtime-recipe/fingerprint implementation and admission tests; `evidence/DGR-003/` |
+| DGR-004 | exact upstream pin, numbered patch series, reproducible fetch/apply/build smoke; `evidence/DGR-004/` |
+| DGR-005 | dense-Llama range ownership loader and memory evidence; `evidence/DGR-005/` |
+| DGR-006 | decode `TensorBundle` protocol amendment, typed tail-result contract, architecture boundary adapter/parity tests and results; `evidence/DGR-006/` |
+| DGR-007 | concurrent session/KV manager, isolation/cleanup tests; `evidence/DGR-007/` |
+| DGR-008 | standalone C++ gRPC worker, fake-model integration tests, lifecycle evidence; `evidence/DGR-008/` |
+| DGR-009 | Meshnet backend/registration/relay integration and tests; `evidence/DGR-009/` |
+| DGR-010 | real local two-process commands, raw metrics and parity report; `evidence/DGR-010/` |
+| DGR-011 | two-machine configuration, commands, hardware/network manifest and raw results; `evidence/DGR-011/` |
+| DGR-012 | continuous scheduler/admission implementation and 1/2/4/8 concurrency report; `evidence/DGR-012/` |
+| DGR-013 | failure/cancel/restart test matrix and resource-cleanup evidence; `evidence/DGR-013/` |
+| DGR-014 | immutable final comparison against DGR-001 thresholds and ship/stop recommendation; `evidence/DGR-014/` |
+| DGR-015 | Qwen3-family adapter, architecture-specific parity/admission/performance evidence; `evidence/DGR-015/` |
+| DGR-016 | narrow upstream patches/tests, design note and human-ready outreach package; `evidence/DGR-016/` |
+| DGR-017 | exact GLM-5.2/GGUF target manifest, resource planner, immutable alpha contract and upstream status; `evidence/DGR-017/` |
+| DGR-018 | verified whole-model `UD-IQ1_S` oracle with native GLM semantic evidence; `evidence/DGR-018/` |
+| DGR-019 | explicit range-owned GLM MoE/MLA/DSA/IndexShare adapter, fixtures and parity; `evidence/DGR-019/` |
+| DGR-020 | real multi-node GLM-5.2 Max target evidence and immutable `alpha`/`stop` verdict; `evidence/DGR-020/` |
+
+## Dependency handoff rule
+
+For every dependency listed by Ralph:
+
+1. Confirm its `passes` state in `prd.json`.
+2. Read `.scratch/distributed-gguf-runtime/evidence/<DEPENDENCY-ID>/README.md`.
+3. Verify referenced source paths and commands still exist.
+4. Do not repeat completed work unless verification exposes a concrete defect.
+5. If dependency evidence is missing or contradictory, stop and repair the dependency instead of guessing.
+
+## Testing and hardware rules
+
+Default tests must be deterministic, GPU-free, model-download-free and API-credit-free.
+
+Real model tests require:
+
+```text
+MESHNET_ENABLE_REAL_INFERENCE_TESTS=1
+```
+
+On this machine:
+
+- Use `.venv-rocm` for real Radeon 8060S ROCm execution.
+- The default Python 3.14 `.venv` is unsuitable for real ROCm inference.
+- Resolve model storage through the machine-specific `.env.<hostname>` configuration.
+- Never download model artifacts under `/home`.
+- Real acceptance must exercise actual Tracker-routed CPU/GPU computation; synthetic workers are only unit tests.
+
+Record exact:
+
+- Model/revision and Artifact hash.
+- Quantization and runtime recipe.
+- Host/hardware/backend/driver.
+- Commands and environment names without secrets.
+- Raw output and metrics.
+- Whether the evidence is synthetic, local-real, or multi-machine-real.
+
+## Worktree and commit discipline
+
+This repository may contain pre-existing changes from research or another feature.
+
+- Inspect `git status` before editing.
+- Never reset, checkout over, stash, delete or reformat unrelated changes.
+- Stage only files belonging to the selected story.
+- Exclude `.ralph-tui`, iteration logs, caches, generated builds, FUSE artifacts and unrelated scratch work.
+- Keep one scoped commit per completed story when the supervising loop requests commits.
+- Do not modify `passes` for another story.
+
+## Mandatory finish/handoff sequence
+
+Before emitting `<promise>COMPLETE</promise>`:
+
+1. Verify every acceptance criterion with real command output or file evidence.
+2. Run story-specific gates and repository quality gates.
+3. Write `.scratch/distributed-gguf-runtime/evidence/<TASK-ID>/README.md` containing:
+   - Summary of changes.
+   - Exact files changed.
+   - Commands run and their real results.
+   - Performance/correctness evidence.
+   - Known limitations and deferred work.
+   - Compatibility or migration notes.
+   - Clear handoff for dependent stories.
+4. Save machine-readable evidence beside it when the story produces metrics or schemas.
+5. Update the source issue status to `done` only after all gates pass.
+6. Preserve failures honestly. Never fabricate model, benchmark, test or hardware output.
+
+## Authoritative references
+
+Active decisions:
+
+- `.scratch/distributed-gguf-runtime/README.md`
+- `.scratch/distributed-gguf-runtime/implementation-strategy.md`
+- `.scratch/distributed-gguf-runtime/architecture.md`
+- `.scratch/distributed-gguf-runtime/ADR-0020-distributed-gguf-runtime.md`
+- `.scratch/distributed-gguf-runtime/PRD.md`
+- `.scratch/distributed-gguf-runtime/prd.json`
+
+Source research:
+
+- `docs/research/distributed-gguf-landscape.md`
+- `docs/research/distributed-gguf-github-followup.md`
+- `docs/research/vllm-distributed-gguf-assessment.md`
+
+If historical notes conflict with these files, the active decisions above win.

.scratch/distributed-gguf-runtime/README.md

@@ -1,63 +1,48 @@
-# Distributed GGUF runtime — planning index
+# Performant concurrent distributed GGUF runtime
 
-Status: draft scratch package.
+Status: active benchmark-gated implementation program.
 
-Goal: make the node network capable of serving large, high-quality open models by distributing GGUF/model artifacts over a torrent-style swarm while executing inference over a sticky multi-node route with per-shard local KV cache.
+## Objective
 
-This scratch supersedes the old assumption in [ADR-0001](../../docs/adr/0001-pytorch-over-llama-cpp.md) that llama.cpp is only a single-node leaf backend. That assumption was correct for the original llama.cpp RPC shape, but the target is now different: torrent-distributed GGUF artifacts plus an explicit route/KV protocol owned by this platform, ideally developed in collaboration with upstream llama.cpp.
+Serve the exact pinned GLM-5.2 `UD-IQ1_S` artifact in `reasoning_effort=max` mode across consumer machines with useful measured performance. Dense Llama is a structural fixture; the real multi-node GLM target is the alpha release gate.
 
-## Artifacts
+See **[GLM-5.2 Max distributed alpha roadmap](GLM-5.2-MAX-ALPHA-ROADMAP.md)** for the target identity, minimum hardware, immutable acceptance matrix, and revised execution order. The 224-GiB figure is an experimental hard-fit floor; recommended topology is 5×64 GiB or 3×96/128 GiB after the required per-node reserve.
 
-| Path | Purpose |
-|---|---|
-| [architecture.md](./architecture.md) | Proposed runtime architecture, data flow, session state, and failure model |
-| [technical-challenges.md](./technical-challenges.md) | Detailed challenge/solution register with acceptance tests |
-| [decision-framework.md](./decision-framework.md) | Grilling framework for open decisions and recommended answers |
-| [research-prior-art.md](./research-prior-art.md) | Prior-art notes for Petals, exo, Distributed Llama, prima.cpp, llama.cpp, DeepSeek-V4-Flash, GLM-5.2, and Ornith |
-| [ADR-0020-distributed-gguf-runtime.md](./ADR-0020-distributed-gguf-runtime.md) | Draft decision record for the GGUF/llama.cpp distributed runtime |
-| [PRD.md](./PRD.md) | Product/runtime requirements and acceptance criteria |
-| [milestones.md](./milestones.md) | Dependency-ordered implementation milestones |
-| [issues/](./issues/) | Implementation-ready tracer-bullet issue briefs |
+## Critical path
 
-## Decision Summary
+```text
+Meshnet control plane
+  -> versioned gRPC/Protobuf Shard protocol
+  -> project-owned standalone C++ worker
+  -> small pinned llama.cpp patch stack
+```
 
-Adopt a hybrid runtime:
+Transformers/safetensors remains the correctness baseline. vLLM remains an optional complete managed provider and a design donor; it is not forked into the public mesh.
 
-- **Weights and artifacts**: distributed by torrent / content-addressed storage / optional CDN.
-- **Hot KV cache**: local to the node that owns the corresponding layer range.
-- **Prefix snapshots**: optionally persisted to cache servers for reuse, retry, and failover.
-- **Active route**: sticky for one request/session.
-- **Context cap**: 128K hard product limit for large models unless explicitly revised.
-- **Backends**: keep PyTorch for fast model-architecture coverage and validation; add llama.cpp/GGUF as the performance path for supported models.
-- **Client feedback**: stream token deltas when feasible; always expose Generation Telemetry.
-- **First serious target model**: DeepSeek-V4-Flash after a smaller GGUF protocol smoke test.
+## Planning artifacts
 
-## What We Learned
+- **[Mandatory Ralph context](RALPH-CONTEXT.md)** — read first in every fresh iteration
+- [Task evidence contract](evidence/README.md)
+- [Implementation strategy](implementation-strategy.md)
+- [Current architecture](architecture.md)
+- [PRD](PRD.md)
+- [Ralph backlog](prd.json)
+- [ADR-0020](ADR-0020-distributed-gguf-runtime.md)
+- [Milestones](milestones.md)
+- [Issues](issues/)
+- [Distributed GGUF research](../../docs/research/distributed-gguf-landscape.md)
+- [GitHub follow-up](../../docs/research/distributed-gguf-github-followup.md)
+- [vLLM assessment](../../docs/research/vllm-distributed-gguf-assessment.md)
 
-- Our current full-model PyTorch path uses Transformers `generate()` and gets local KV cache.
-- Our current distributed PyTorch path disables cache and recomputes the full growing sequence per token.
-- The seam today carries hidden activations, not KV cache; at 128K this becomes impossible for serious models if repeated every decode token.
-- The missing capability is not "send KV across the network"; it is **stable per-session local KV cache per shard**.
-- GGUF distribution is solved enough at the artifact layer, but GGUF/llama.cpp needs explicit layer-boundary execution APIs for our route model.
+## Ralph execution
 
-## Recommended Order
+Use supervised one-story iterations for this high-risk runtime:
 
-See [milestones.md](./milestones.md) for the full dependency map.
+```bash
+ralph-tui run \
+  --prd .scratch/distributed-gguf-runtime/prd.json \
+  --agent claude --model opus \
+  --iterations 1 --no-tui --no-setup --verify
+```
 
-1. [01 — Route Session lifecycle](./issues/01-route-session-lifecycle.md)
-2. [02 — Prefill/decode binary HTTP protocol](./issues/02-prefill-decode-binary-http.md)
-3. [03 — Generation Telemetry and streaming response contract](./issues/03-generation-telemetry-and-streaming.md)
-4. [04 — PyTorch distributed KV reference route](./issues/04-pytorch-distributed-kv-reference.md)
-5. [05 — Local llama.cpp/GGUF backend](./issues/05-local-llamacpp-gguf-backend.md)
-6. [06 — Model Artifact manifest and Shard advertisement](./issues/06-model-artifact-manifest.md)
-7. [07 — llama.cpp layer-boundary prototype](./issues/07-llamacpp-layer-boundary-prototype.md)
-8. [08 — Networked distributed GGUF route](./issues/08-networked-distributed-gguf-route.md)
-9. [09 — DeepSeek-V4-Flash support audit](./issues/09-deepseek-v4-flash-support-audit.md)
-10. [10 — GLM-5.2 and Ornith follow-up support audit](./issues/10-glm52-ornith-followup-audit.md)
-
-## Open Questions
-
-- Does upstream llama.cpp already expose enough internal API for arbitrary layer-range execution and hidden-state boundary I/O, or do we need an extension?
-- Can GGUF split metadata be made layer/tensor semantic enough for torrent placement and partial loading?
-- What is the minimum protocol needed for compressed KV formats such as GLM-5.2 DSA/MLA without exposing model-specific internals to the tracker?
-- How much reliability do we need in alpha: fail request on route loss, or support route repair with KV snapshots?
+Inspect the diff, run the story gates, and commit one verified story before the next iteration. Real-model stories require the explicit environment gate and mounted-drive model storage.

.scratch/distributed-gguf-runtime/architecture.md

@@ -1,274 +1,264 @@
-# Distributed GGUF Runtime Architecture
+# Performant Concurrent Distributed GGUF Architecture
 
-## Product Stance
+Status: current target architecture
+Last updated: 2026-07-13
 
-The platform optimizes for access to high-quality models, not lowest latency. Latency is acceptable if the user can run models that are otherwise unavailable to them. The hard context limit for the first serious distributed runtime should be **128K tokens**. Longer context usually means the product is compensating for missing task decomposition, retrieval, or workspace summarization.
+## Product invariant
 
-## Current State
+The system exists to serve high-quality models that exceed one consumer node's memory while retaining useful interactive speed and aggregate concurrency. A feature that only produces a distributed demo but is slower, globally serialized, or impossible to operate on consumer hardware is not complete.
 
-The current node has two materially different inference paths:
+The alpha target is the exact pinned GLM-5.2 `UD-IQ1_S` artifact in `reasoning_effort=max` mode. Its target-specific architecture/resource/acceptance contract is [GLM-5.2-MAX-ALPHA-ROADMAP.md](GLM-5.2-MAX-ALPHA-ROADMAP.md). Dense Llama is a structural fixture, not the product target.
 
-- **Full local PyTorch model**: calls Hugging Face `model.generate()`, so Transformers owns autoregressive decode and local KV cache.
-- **Distributed PyTorch route**: bypasses `model.generate()`, calls individual layers with `use_cache=False`, and recomputes the full growing sequence for every generated token.
+## Existing control plane
 
-Current distributed data flow:
+Meshnet remains the only public control plane:
+
+- Tracker registration, Coverage Map, route scoring and assignment.
+- Contiguous Shards and overlap-safe effective starts.
+- Stable Route Sessions and route epochs.
+- Local per-Shard Hot KV State in the reference backend.
+- Direct/relay transport, cancellation and backpressure.
+- Generation Telemetry, billing, validation and per-node attribution.
+- Model-agnostic capability admission.
+
+No external engine replaces these responsibilities.
+
+## Runtime topology
 
 ```text
-client request
-  -> head node formats prompt
-  -> for each output token:
-       head tokenizes full current text
-       head runs early layers over all tokens
-       head sends full activation [batch, sequence, hidden] to next node
-       middle nodes run their layers over all tokens
-       tail returns one decoded token string
-       head appends token to text
+OpenAI-compatible client
+        |
+Gateway / Tracker Node
+        |
+ordered Inference Route
+        |
+        +-- head Shard: tokenizer/embedding + early layers
+        |       local weights and Hot KV State
+        |
+        +-- middle Shard(s): architecture boundary + owned layers
+        |       local weights and Hot KV State
+        |
+        +-- tail Shard: final layers + norm/head/sampling
+                local weights and Hot KV State
 ```
 
-This is correct for small demos but not viable for large models. For GLM-5.2, a single 128K seam activation is roughly:
+Weights never move in the per-request hot path. Every node opens and verifies its local Model Artifact before becoming routable.
+
+## Primary execution substrate
 
 ```text
-128K tokens * hidden_size 6144 * 2 bytes ~= 1.5 GiB per hop
+project-owned C++ Shard worker
+        |
+small exact-commit llama.cpp patch stack
+        |
+GGUF mmap, quantized kernels, architecture graphs,
+KV/sequence operations, CPU/CUDA/HIP/Vulkan/Metal backends
 ```
 
-Sending that every output token is the bottleneck.
+The patch stack adds only the missing local execution seam:
 
-## Target State
+1. Range-aware tensor registration/loading.
+2. Endpoint-specific embedding and final head ownership.
+3. Architecture-defined intermediate input.
+4. Architecture-defined pre-tail boundary output.
+5. Layer-filtered KV and external session mapping.
 
-Target distributed data flow:
+The worker owns protocol translation and process lifecycle. llama.cpp never receives Tracker, relay, billing or volunteer-network code.
+
+## Shard data plane
+
+Use Protocol Buffers and gRPC over HTTP/2.
+
+### Service shape
+
+- Unary capability and health.
+- Bidirectional Route Session stream.
+- Explicit release and cancellation.
+- Metrics suitable for capability admission and route scoring.
+
+### Session stream
+
+One long-lived stream represents one Route Session Activation Seam. It amortizes connection setup and inherits HTTP/2 flow control. Every message carries enough identity to reject stale or incompatible work.
 
 ```text
-client request
-  -> tracker selects route and pins session
-  -> head node creates session_id
-  -> prefill:
-       prompt is chunked
-       each shard computes its layer range
-       each shard appends local KV/state for its own layers
-       activations cross only layer seams
-  -> decode loop:
-       head sends one new token / one-step hidden state
-       each shard reads local KV/state for session_id
-       each shard appends one step to local KV/state
-       only one-step activation crosses seams
-       tail returns logits/token
+schema version
+request/work id
+Route Session id
+route epoch
+Model Artifact hash
+runtime recipe fingerprint
+Shard begin/end and effective start
+prefill/decode/release/cancel phase
+position and token range
+idempotency step id
+cache expectation/result
+named tensor bundle
+compression/checksum
 ```
 
-The KV cache remains local to the node that computed it. It is not sent to the next node and not read from a remote cache server during every decode step.
+Prefill tensors are split into bounded ordered frames. Decode messages carry one-step architecture boundary bundles and remain small. DGR-006 amends the current v1 decode fast path—which carries only one `NamedTensor`—to carry a versioned `TensorBundle`, while preserving compact one-tensor encoding and explicit compatibility behavior.
 
-## Client Feedback
+Tail completion is not inferred from an activation tensor name. The protocol exposes a typed logits and/or sampled-token result, and exact sampling parameters plus chat-template/reasoning mode are bound to request/runtime identity.
 
-Streaming responses are desirable when the backend and client transport support them. The product should stream token deltas when possible, and it must always provide realtime Generation Telemetry while the route is working.
+Direct nodes use gRPC. Nodes requiring the existing relay carry the same protobuf frames as opaque binary through the relay session. This preserves one semantic protocol instead of maintaining separate direct and relay payload contracts.
 
-The fallback behavior is a non-streaming final answer plus live telemetry. That fallback is acceptable for early route proofs or models/backends that cannot expose clean token deltas yet, but the preferred client experience is streamed output plus telemetry.
+## Architecture boundary
 
-Minimum client-visible telemetry:
-
-- route/session accepted
-- selected model and quantization
-- prefill phase started/completed
-- decode phase started
-- generated token count
-- rolling tokens per second
-- route health or retry/failure reason
-- estimated billing units when available
-
-Implementation options:
-
-- Server-Sent Events or WebSocket for realtime progress
-- polling endpoint for simple clients
-- OpenAI-compatible streaming for clients that require token deltas
-
-This means "no token streaming" is acceptable only as a fallback. "Silent wait for minutes" is not acceptable.
-
-## Artifact Plane
-
-Artifact distribution is separate from execution.
+The public boundary is a versioned named-tensor bundle:
 
 ```text
-model publisher
-  -> produces model manifest
-  -> creates GGUF / safetensors / tokenizer artifacts
-  -> content-addresses every file/chunk
-  -> publishes torrent/magnet + HTTP fallback metadata
-
-node
-  -> chooses model/layer range
-  -> downloads needed files/chunks
-  -> verifies hash
-  -> advertises availability to tracker
+bundle schema/version
+architecture adapter and boundary point
+named tensors
+per-tensor shape, dtype and byte order
+payload fragments
+compression/checksum
 ```
 
-Required manifest fields:
+Dense Llama may use one residual tensor. Other adapters may require more. vLLM's Llama and Qwen3-MoE PP paths demonstrate a boundary with both `hidden_states` and `residual`; therefore the generic protocol must not assume one anonymous tensor.
 
-- model id and version
-- upstream source repo and revision
-- license
-- architecture name
-- tokenizer files and hashes
-- quantization
-- tensor-to-layer map
-- file/chunk hashes
-- optional GGUF split files
-- supported runtime backends
-- context cap
-- KV/cache format descriptor
+GLM-5.2 normally exchanges a 6,144-element hidden state. If a memory-balanced Shard boundary splits an IndexShare Full producer from Shared consumers, the bundle also carries the typed top-k index sideband. The planner prefers boundaries that keep an IndexShare ownership group local, but the protocol validates the sideband rather than assuming it never crosses a seam.
 
-## Execution Plane
+Only the head owns token embedding. Only the tail owns final normalization, LM head and sampling. Middle Shards exchange the architecture-defined pre-tail boundary, not final normalized embeddings.
 
-The tracker selects routes using layer coverage and observed performance:
+## Hot KV State and concurrency
 
 ```text
-route = [
-  head node: embeddings + layers 0..k
-  middle nodes: contiguous layer ranges
-  tail node: final layers + norm + lm_head
-]
+(Route Session id, route epoch)
+       -> local llama sequence or bounded context
+       -> KV for owned layers only
+       -> lease, memory accounting and lifecycle
 ```
 
-Route selection inputs:
+Required operations:
 
-- model id/version/quantization
-- layer coverage
-- node hardware
-- measured prefill throughput
-- measured decode throughput
-- queue depth
-- latency to neighboring nodes
-- cache warmth for the requested prefix/session
-- reliability/reputation
+- Prefill append.
+- Decode append.
+- Truncate after rejected speculative positions if later enabled.
+- Explicit release.
+- TTL/LRU eviction.
+- Cache-miss response.
+- Stale-epoch rejection.
 
-The route is sticky for the request/session. A new route means either a fresh prefill or restoring compatible KV snapshots.
+A node must not clear global KV on a new stream or serialize all requests behind one logical serving sequence.
 
-## KV Cache Ownership
+## Continuous batching
 
-KV/state ownership is by layer range:
+Autoregressive dependencies remain sequential inside one Route Session. Aggregate throughput comes from batching compatible decode steps across active sessions:
 
 ```text
-session_id = request scoped id
-node A owns layers 0..15 KV for session_id
-node B owns layers 16..31 KV for session_id
-node C owns layers 32..77 KV for session_id
+time 0: session A token 1 + session B token 8 + session C token 3
+          -> one llama batch for this Shard
+
+time 1: next ready positions from active sessions
+          -> next llama batch
 ```
 
-The tracker does not own hot KV. It may know which nodes hold active KV for session accounting and failure handling.
+The node scheduler:
 
-Cache servers may store:
+- Admits work against weight, KV, scratch and queue budgets.
+- Keeps per-session token positions and outputs separate.
+- Prevents long prefill from starving decode.
+- Applies bounded backpressure.
+- Reports active sessions, queue depth, batch occupancy, KV pressure and throughput.
 
-- prompt-prefix snapshots
-- session checkpoints for retry
-- cold reusable context blocks
-- audit samples
+The initial deterministic gate is four concurrent sessions on a small model without cross-talk. Hardware-specific limits are measured and advertised through capability admission.
 
-Cache servers must not be in the per-token hot loop unless colocated with the compute node.
+## Parallelism boundaries
 
-## 128K KV Budget
+| Mechanism | First-runtime use |
+|---|---|
+| Layer/pipeline parallelism | Public Inference Route across contiguous Shards |
+| Continuous batching | Inside every node across active Route Sessions |
+| Data parallelism | Multiple complete routes for independent requests |
+| Tensor parallelism | Deferred to a trusted composite node/managed cluster |
+| Expert parallelism | Deferred to a trusted composite node/managed cluster |
+| Disaggregated prefill | Deferred until core route performance passes |
+| Speculative decoding | Deferred optimization |
 
-GLM-5.2 compressed DSA/MLA-style estimate from config:
+Public WAN tensor/expert collectives are rejected for the first runtime because their per-layer communication and static rank assumptions conflict with heterogeneous volunteer nodes.
 
-```text
-layers = 78
-kv_lora_rank = 512
-qk_rope_head_dim = 64
-dtype = bf16 = 2 bytes
-context = 128K
+## Optional providers
 
-per_token ~= 78 * (512 + 64) * 2 = 89,856 bytes ~= 87.75 KiB
-128K total ~= 10.7 GiB
-per layer ~= 137 MiB
-```
+### Transformers/safetensors
 
-This is feasible when sharded:
+Remains:
 
-| Layer count | Approx active KV at 128K |
-|---:|---:|
-| 1 | 137 MiB |
-| 10 | 1.37 GiB |
-| 20 | 2.75 GiB |
-| 78 | 10.7 GiB |
+- Correctness/reference backend.
+- Fallback for unsupported architectures.
+- Baseline for performance and output quality.
 
-The exact runtime value depends on implementation and cache quantization, but the order of magnitude is acceptable.
+### vLLM
 
-## Protocol Sketch
+May run unmodified as a complete model or managed TP/PP/EP cluster represented as one logical provider. Its internal ranks are not independently routed or rewarded.
 
-### Prefill
+Borrow only concepts such as named bundles, continuous batching, typed compatibility fingerprints, explicit transfer lifecycle and load telemetry.
 
-```http
-POST /v1/sessions/{session_id}/prefill
-Content-Type: application/octet-stream
-X-Meshnet-Model: zai-org/GLM-5.2
-X-Meshnet-Route-Id: ...
-X-Meshnet-Token-Range: 0-2047
-X-Meshnet-Shape: 1,2048,6144
-X-Meshnet-Dtype: bfloat16
+### Whole-model llama.cpp
 
-<activation bytes>
-```
+Provides a local proxy backend, correctness oracle and performance baseline. It is not the native distributed milestone.
 
-The receiver:
+## Artifact and recipe compatibility
 
-- validates route/session
-- runs assigned layer range for that chunk
-- appends local KV/state
-- forwards resulting activation to next hop
+A routable recipe identifies separately:
 
-### Decode
+- Source Model Artifact hash and optional derivative/slice hash.
+- Architecture and adapter version.
+- Tokenizer revision and vocabulary.
+- Weight quantization.
+- Activation interchange dtype/schema.
+- Backend compute dtype and backend implementation.
+- KV dtype/layout.
+- RoPE/context parameters.
+- llama.cpp commit and project patch version.
+- Shard range and endpoint ownership.
 
-```http
-POST /v1/sessions/{session_id}/decode-step
-Content-Type: application/octet-stream
-X-Meshnet-Model: zai-org/GLM-5.2
-X-Meshnet-Position: 131072
-X-Meshnet-Shape: 1,1,6144
-X-Meshnet-Dtype: bfloat16
+Compatibility fails closed. Similar quantization labels or model names are not enough.
 
-<one-step activation bytes>
-```
+## Admission and failure
 
-The receiver:
+A recipe becomes routable only after a real local and distributed forward passes. Synthetic tests remain unit coverage.
 
-- loads local KV/state by `session_id`
-- runs one decode step for assigned layers
-- appends one token position to local KV/state
-- forwards one-step activation
+Alpha failure behavior:
 
-## GGUF / llama.cpp Integration
+- Deadline or node loss cancels the Route Session.
+- Every node releases KV and queued buffers.
+- Uncertain mutations are not replayed silently.
+- Retry starts from token zero on a newly compatible route.
+- No cross-node KV import is trusted until a later signed/compatible snapshot protocol exists.
 
-The target llama.cpp integration needs more than `llama-server`.
+## Performance release contract
 
-Required capabilities:
+Before native development proceeds, compare the current Transformers/safetensors backend with whole-model llama.cpp under controlled model/hardware/quality lanes.
 
-- load full GGUF locally for immediate single-node performance
-- optionally load only selected tensors/layers
-- execute a layer range against inbound hidden states
-- expose outbound hidden states at a boundary
-- own per-session KV/state for only the loaded layer range
-- support prefill chunks and decode-step calls
-- expose model-specific cache metadata for DSA/MLA without requiring the tracker to understand tensor internals
+Final release compares distributed GGUF with distributed safetensors using thresholds locked before seeing final results.
 
-If llama.cpp cannot expose these as stable APIs today, the collaboration target is an upstream extension rather than a long-lived fork.
+Required measurements:
 
-## Failure Model
+- TTFT.
+- Prefill and decode tokens/sec.
+- Aggregate concurrency throughput.
+- p50/p95 latency.
+- Seam bytes and latency.
+- Queue/batch occupancy.
+- RSS, VRAM and KV pressure.
+- Output-quality drift.
+- Cancellation/failure cleanup.
 
-Alpha behavior:
+The GGUF path ships only if it is faster at acceptable quality or enables a larger otherwise-unroutable model at useful measured speed.
 
-- Route node drops during prefill: fail request and retry from scratch.
-- Route node drops during decode: fail request unless a recent KV snapshot exists.
-- Tracker restart: active sessions may be lost; completed billing records persist.
-- Node restart: local hot KV is lost.
+## Implementation sequence
 
-Later behavior:
+1. Preserve completed DGR-001 performance and DGR-002 protocol contracts.
+2. DGR-017 locks exact GLM-5.2 Max artifact, resource, and alpha acceptance identity.
+3. Define exact recipe identity and pin one reproducible llama.cpp boundary.
+4. Run two lanes in parallel: DGR-018 establishes the whole-model `UD-IQ1_S` oracle on 224+ GiB usable memory, while DGR-005/DGR-006 implement range loading and named boundary parity with a cheap dense fixture.
+5. DGR-019 adds explicit GLM-5.2 MoE/MLA/DSA/IndexShare semantics after both lanes pass.
+6. Implement local KV; build and integrate the standalone worker.
+7. Pass local two-process and real two-physical-machine execution.
+8. Harden cancellation, node loss, restart, and cleanup required by alpha.
+9. DGR-020 executes the exact multi-node target and emits immutable `alpha` or `stop`.
+10. Post-alpha: continuous batching, final comparison, longer context, MTP, and package optimization.
+11. Prepare narrow upstream patches/tests; add Qwen as later architecture expansion.
 
-- periodic KV snapshots for long sessions
-- prefix cache reuse across requests
-- route repair when a semantically equivalent node has the same model/layer range and compatible cache snapshot
-
-## Security And Trust
-
-Activation/KV data can reveal user prompts. Public volunteer routes are not private. For sensitive workloads:
-
-- use private swarms
-- allow paid trusted nodes
-- encrypt transport
-- avoid storing hot KV on untrusted shared cache servers
-- sample outputs for fraud/audit as already planned in alpha hardening
+See [the Ralph backlog](prd.json) and [implementation strategy](implementation-strategy.md).

.scratch/distributed-gguf-runtime/decision-framework.md

@@ -1,5 +1,7 @@
 # Distributed GGUF Decision Framework
 
+> **Superseded for active implementation decisions.** The grill was resolved on 2026-07-13. Use [implementation-strategy.md](implementation-strategy.md), [architecture.md](architecture.md), [ADR-0020](ADR-0020-distributed-gguf-runtime.md), and [prd.json](prd.json). This file remains as historical decision rationale.
+
 This framework is for grilling open decisions. It keeps decisions tied to project vocabulary and implementation gates instead of vague "distributed inference" language.
 
 ## Core Vocabulary

.scratch/distributed-gguf-runtime/evidence/DGR-001/BLOCKED.md

@@ -0,0 +1,43 @@
+# DGR-001 downstream stop-condition handoff
+
+Status: **DGR-001 is complete; native-track promotion is blocked by the immutable v1 verdict.**
+
+This is no longer an execution-prerequisite blocker. The required real benchmark
+ran successfully, every recipe completed at concurrency 1 and 4, artifacts were
+verified, and deterministic/full test gates passed.
+
+## Locked result
+
+`contract-evaluation.json` records:
+
+```text
+verdict: stop
+quality_lane_pass: false
+speed_benefit: true
+fit_benefit: true
+stop_condition_met: true
+```
+
+The exact-revision BF16 GGUF quality lane compared every prompt but achieved
+`0.3333` exact match and `0.9471` mean similarity against the Transformers BF16
+reference. V1 requires `0.90` and `0.97`. Quantized Q4_K_M had substantial speed
+and fit benefits, but the contract explicitly forbids speed from redeeming a
+failed near-lossless quality lane.
+
+## Scope of this stop
+
+The measured baseline is Qwen2.5-0.5B on CPU using a CPU-only llama.cpp build.
+It is not a Radeon, large-model, distributed, or native-shard result. Therefore:
+
+1. Do not silently mark v1 promoted or weaken its thresholds after observing the
+   data.
+2. Do not let DGR-004 or later runtime stories treat DGR-001 completion as a
+   positive promotion signal.
+3. A human may choose one of these explicit paths:
+   - stop the native GGUF track as v1 directs;
+   - diagnose and fix the BF16 runtime divergence, then rerun the exact v1 plan;
+   - authorize a separately versioned GPU/large-model contract whose scope and
+     workload are locked before its measurements.
+
+All raw evidence, configuration, artifacts, hashes, and reproduction commands
+are in this directory and `README.md`.

.scratch/distributed-gguf-runtime/evidence/DGR-001/README.md

@@ -0,0 +1,199 @@
+# DGR-001 — Safetensors versus GGUF performance contract
+
+Status: **complete; immutable v1 verdict is `stop`.**
+
+DGR-001 successfully produced a controlled local-real CPU baseline. Completion
+means the experiment and decision contract are durable and verified; it does
+**not** mean the native GGUF track is approved to continue. The locked quality
+gate failed, so dependent runtime work requires a human decision or a new,
+explicitly versioned experiment/contract rather than silently weakening v1.
+
+## Controlled workload
+
+- Model: `Qwen/Qwen2.5-0.5B-Instruct`
+- Exact source revision: `7ae557604adf67be50417f59c2c2f167def9a775`
+- Machine: `fedora`, Linux `7.0.14-101.fc43.x86_64`, 32 logical CPUs
+- Device: CPU for every recipe; VRAM is therefore correctly reported as zero
+- Runtime reference: Transformers `5.13.0`, PyTorch
+  `2.10.0+rocm7.13.0a20260513`, BF16 safetensors
+- GGUF runtime: llama.cpp version 9991, commit
+  `e920c523e3b8a0163fe498af5bf90df35ff51d25`
+- Workload: three fixed short/medium/long prompts, greedy sampling, 32 output
+  tokens, three repeats, two warmups, concurrency 1 and 4, 16 CPU threads
+- Evidence class: `local-real`
+
+All artifacts are beneath `/run/media/popov/DATA/llm/`; no model artifact was
+created under `/home`.
+
+## Recipes and exact artifacts
+
+| Recipe | Artifact | SHA-256 |
+|---|---|---|
+| Transformers BF16 reference | complete mounted Hugging Face snapshot | `e596e9d6205fdc9177569cccd7f8b471b058f66e3630c8e4326d5aad52bd18b6` |
+| llama.cpp BF16 quality lane | `Qwen2.5-0.5B-Instruct-7ae5576-BF16.gguf` | `e842fdc35d7f00fda95a54e1b51731ba1d196aea45065cc9f46925fdc1d6f862` |
+| llama.cpp Q4_K_M performance/fit lane | `Qwen2.5-0.5B-Instruct-7ae5576-Q4_K_M.gguf` | `a88e3f570e2efeaf06b50df9859db2c70d8646aa3a2c94a14e14d5797a2921a5` |
+
+The snapshot digest covers every sorted relative path, resolved size, and file
+byte, so tokenizer/config drift is included. The BF16 GGUF was converted
+directly from the exact snapshot while preserving BF16 weights. Q4_K_M was
+quantized from an exact-revision F16 conversion with the pinned quantizer.
+Runtime validation recomputes every declared digest before model loading.
+
+## Real results
+
+All recipes completed every request with zero failures.
+
+| Metric | Transformers BF16 | llama.cpp BF16 | llama.cpp Q4_K_M |
+|---|---:|---:|---:|
+| Decode tok/s, c=1 | 40.8 | 98.5 | 207.7 |
+| Aggregate decode tok/s, c=4 | 46.5 | 222.8 | 195.7 |
+| TTFT p50, c=1 | 40.0 ms | 15.1 ms | 21.6 ms |
+| Peak resident memory, c=1 | 1.94 GB | 1.11 GB | 0.54 GB |
+| Artifact size | 1.00 GB | 0.99 GB | 0.40 GB |
+| Failures | 0 | 0 | 0 |
+
+Against the reference, the eligible Q4_K_M lane measured:
+
+- single-request decode speedup: **5.10×**;
+- concurrency-4 aggregate throughput speedup: **4.20×**;
+- resident-memory ratio: **0.279×**;
+- artifact-size ratio: **0.398×**.
+
+The near-lossless BF16 quality lane compared all three prompts but measured:
+
+- exact match: **0.3333** (v1 requires at least `0.90`);
+- mean text similarity: **0.9471** (v1 requires at least `0.97`).
+
+Tokenization and stopping were controlled: every runtime saw the same prompt
+token counts and reported 31 post-TTFT decode tokens. The v1 mismatch is a
+real greedy-output divergence on two prompts, not missing coverage or a
+text-length artifact. Its root cause remains undetermined; no post-contract
+logit-tie claim is acceptance evidence. Therefore `contract-evaluation.json`
+records:
+
+```text
+verdict: stop
+quality_lane_pass: false
+speed_benefit: true
+fit_benefit: true
+stop_condition_met: true
+```
+
+Thresholds were not changed after observing these results.
+
+## Post-contract parity and ROCm diagnostics
+
+`summarize-quality-parity.py` verifies and separates two signed sources. The CPU
+v1 row uses CPU kernels and a Transformers BF16 oracle; it remains at `0.3333`
+exact match with an unexplained divergence. The ROCm row uses a different plan,
+GPU kernels, and a Transformers float32 oracle. In that narrower diagnostic,
+the same BF16 GGUF artifact matches all three 32-token sequences exactly (`1.0`
+exact match and `1.0` similarity). No conversion corruption was observed in
+that three-sequence ROCm sample; this does not prove global conversion
+correctness or explain the CPU result.
+
+A separate HIP build at commit `e920c523` was compiled for `gfx1151` and
+measured `ROCm0: Radeon 8060S Graphics`; its `llama-server` SHA-256 is
+`b6bb4da687dbde86e243ba006cef05919b7b97255cd7e2371e1d451220aca139`.
+A signed `gpu-diagnostic` profile measured zero failures:
+
+| GPU metric | Transformers BF16 ROCm | llama.cpp Q4 ROCm | Q4 ratio |
+|---|---:|---:|---:|
+| Decode tok/s, c=1 | 81.12 | 251.25 | **3.10×** |
+| Aggregate decode tok/s, c=4 | 91.24 | 511.33 | **5.60×** |
+| TTFT p50, c=1 | 13.77 ms | 11.80 ms | **0.857×** |
+
+The GPU report is signed under the distinct
+`run_configured_gpu_diagnostic/v1` producer. The v1 evaluator rejects that
+producer even when its signature is valid. llama-server process VRAM remains
+unmeasured, so this diagnostic cannot replace or satisfy the immutable v1
+contract. Its signed backend detail records the measured `ROCm0: Radeon 8060S
+Graphics` device and `25/25` offloaded layers.
+
+## Implementation
+
+- `recipe_benchmark.py` provides the runtime-neutral measurement core, true
+  concurrency, continuous in-flight peak-memory sampling, percentile/throughput
+  aggregation, failures, and output drift.
+- `recipe_drivers.py` provides opt-in Transformers and llama-server drivers,
+  mounted-drive confinement, exact artifact/runtime verification, equal
+  device/thread budgets, greedy-only validation, measured host provenance, a
+  CPU-only v1 guard until process VRAM can be measured honestly, and a distinct
+  signed GPU diagnostic profile that the v1 evaluator cannot accept.
+- Peak RSS is runtime-scoped: Transformers reports growth above its pre-runtime
+  Python baseline, while llama.cpp reports its isolated server process tree.
+  Both are sampled continuously during in-flight requests.
+- TTFT uses each runtime's prompt/first-token compute boundary; end-to-end HTTP,
+  scheduling, and queue overhead remains in latency and `queue_wait_ms`.
+- The exact canonical plan SHA-256 locks prompts, model/revision, sampling,
+  output length, repeats, warmups, and concurrency. The evaluator also requires
+  equal prompt/decode token counts across recipes.
+- llama.cpp's `predicted_n` includes the first token while `predicted_ms` begins
+  after it; the driver subtracts that token so decode throughput matches the
+  Transformers inter-token convention.
+- `performance_contract.py` rejects wrong plans, unsigned or incorrectly signed
+  real evidence, wrong config/artifact/runtime/backend/host bindings, missing
+  recipes/concurrency, mixed model revisions, incomplete quality coverage, and
+  failed references.
+- Every non-synthetic report is Ed25519-signed over the complete canonical JSON,
+  including raw outcomes and metrics. The contract pins the public key and exact
+  config SHA-256; the private key remains outside Git at mode `0600`.
+- The signer fingerprint is independently anchored outside this evidence
+  directory in `../../trusted-evidence-signers.json` and checked by tests.
+- Quantized drift remains advisory. Only the near-lossless lane can satisfy the
+  quality gate, and only performance-fit recipes can earn speed/fit benefits.
+
+## Evidence files
+
+- `performance-contract.json` — immutable v1 thresholds and stop condition
+- `benchmark-config.json` — exact real-run plan, drivers, artifacts, and hashes
+- `results.json` — raw machine-readable per-request and aggregate evidence
+- `results.txt` — human-readable benchmark summary
+- `baseline.json` — distilled measurements for later comparison
+- `contract-evaluation.json` — fail-closed v1 verdict
+- `quality-parity-diagnosis.json` / `.md` — run/device-scoped signed-evidence summary
+- `summarize-quality-parity.py` — verifies both evidence chains and regenerates it
+- `gpu-diagnostic-config.json` — exact ROCm diagnostic artifacts and runtimes
+- `gpu-diagnostic-results.json` / `.txt` — signed GPU outcomes and summary
+- `commands.txt` — reproducible conversion, benchmark, evaluation, and test commands
+- `BLOCKED.md` — downstream stop-condition handoff
+- `known-unrelated-failure.md` — clean-base reproduction of the tracker race
+- `../../trusted-evidence-signers.json` — repository-reviewed signer fingerprint
+
+## Verification
+
+```text
+Targeted: 28 passed (5/5 consecutive focused runs)
+Latest full suite: 755 passed, 13 skipped
+Earlier full suite: 751 passed, 13 skipped
+Current cancellation retry matrix, DGR-001: 4/5 passed
+Earlier cancellation retry matrix, clean d904c40: 4/5 passed
+compileall: passed
+git diff --check: passed
+Evidence JSON parse/integrity checks: passed
+```
+
+The intermittent tracker cancellation race reproduced at the same rate on the
+clean base and is retained in `known-unrelated-failure.md`; the final full suite
+completed green. DGR-001 changes no tracker/proxy files.
+
+The earlier Ralph claim that the full suite was blocked by Protobuf 6.33.6 was
+invalid: it used Hermes Agent's internal venv. Verification above used the
+project `.venv`, which has the DGR-002-compatible runtime. Real inference used
+`.venv-rocm` Python 3.12.
+
+## Limitations and dependent-story handoff
+
+- The immutable contract result is a **0.5B CPU baseline**. The separate Radeon
+  diagnostic is real local GPU evidence, but neither result covers a large
+  model, distributed execution, network transport, or a native shard worker.
+- A separate `GGML_HIP=ON` llama.cpp build exists and produced GPU timings, but
+  llama-server process VRAM is not measurable by the current driver; GPU
+  memory/fit claims therefore remain ineligible for v1.
+- Absolute timings are developer-machine measurements; locked ratios and raw
+  artifacts are provided for reproducibility.
+- DGR-014 may consume v1 only with the exact plan/evidence requirements enforced
+  by `performance_contract.py`.
+- DGR-004 and later native-runtime work must not treat DGR-001 completion as a
+  promotion. V1 says `stop`; proceeding requires a human decision backed by a
+  separately versioned GPU/large-model contract or a diagnosed quality fix.

.scratch/distributed-gguf-runtime/evidence/DGR-001/baseline.json

@@ -0,0 +1,169 @@
+{
+  "artifact_sha256": {
+    "llama-cpp-near-lossless-quality": "e842fdc35d7f00fda95a54e1b51731ba1d196aea45065cc9f46925fdc1d6f862",
+    "llama-cpp-quantized-performance-fit": "a88e3f570e2efeaf06b50df9859db2c70d8646aa3a2c94a14e14d5797a2921a5",
+    "transformers-safetensors-reference": "e596e9d6205fdc9177569cccd7f8b471b058f66e3630c8e4326d5aad52bd18b6"
+  },
+  "backend_detail": {
+    "llama-cpp-near-lossless-quality": "version: 9991 (e920c523) | built with GNU 15.2.1 for Linux x86_64; binary sha256 fd8fe612970f23e447f2e717cfa51665be06b8d7315ba60556e010f6bca510dd; threads 16; parallel slots 4; ctx/slot 512; gpu layers 0",
+    "llama-cpp-quantized-performance-fit": "version: 9991 (e920c523) | built with GNU 15.2.1 for Linux x86_64; binary sha256 fd8fe612970f23e447f2e717cfa51665be06b8d7315ba60556e010f6bca510dd; threads 16; parallel slots 4; ctx/slot 512; gpu layers 0",
+    "transformers-safetensors-reference": "torch 2.10.0+rocm7.13.0a20260513; dtype bfloat16; device cpu; intra-op threads 16"
+  },
+  "evidence_class": "local-real",
+  "host": {
+    "accelerator_name": "Radeon 8060S Graphics",
+    "accelerator_runtime": "7.13.26183",
+    "benchmark_lane": "cpu-controlled-baseline",
+    "converter_sha256": "c819f18fb22927b49fabc3b35d1c9e21ee638b3817eccd1bd4efbcc7116eeb4d",
+    "cpu_count": 32,
+    "cuda_available": true,
+    "hostname": "fedora",
+    "llama_cpp_commit": "e920c523e3b8a0163fe498af5bf90df35ff51d25",
+    "llama_cpp_version": "9991",
+    "llama_server_identities": {
+      "/run/media/popov/d/DEV/llamacpp/llama.cpp/build/bin/llama-server": {
+        "sha256": "fd8fe612970f23e447f2e717cfa51665be06b8d7315ba60556e010f6bca510dd",
+        "version": "version: 9991 (e920c523) | built with GNU 15.2.1 for Linux x86_64"
+      }
+    },
+    "llama_server_sha256": "fd8fe612970f23e447f2e717cfa51665be06b8d7315ba60556e010f6bca510dd",
+    "platform": "Linux-7.0.14-101.fc43.x86_64-x86_64-with-glibc2.42",
+    "python": "3.12.13",
+    "quantizer_sha256": "bd0cc8c7be6d48aad4755b31062e0e59a887cbadd43dbb8771853d5858bb198f",
+    "torch_version": "2.10.0+rocm7.13.0a20260513",
+    "transformers_version": "5.13.0"
+  },
+  "model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+  "model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
+  "plan_sha256": "efe24690a9a7164bac6ab3fd0a6b22f078fc08aaefcfb96210ddf154e6050570",
+  "provenance": {
+    "completed_at": "2026-07-13T16:27:19.647692Z",
+    "config_sha256": "00b2cce3e2f281bdf92fc5304ba5cac915a178ffccd3b9a25995ce39c00b90d3",
+    "producer": "meshnet_node.recipe_drivers.run_configured_benchmark/v1",
+    "run_id": "e4eedadf-22f6-4907-8990-985456961099",
+    "schema_version": 1,
+    "signature": "owev+/ToswP20C923G6E+srOCUBV5vrjmndVatr9CbTXakiFGqlHrTiEo+aymA4BcSwmG6KJTxlxO6WpLnpcAg==",
+    "signature_algorithm": "ed25519",
+    "signer_public_key_sha256": "8baca8742d9b3ed0c3fc54929c23f75ec8c1c739900aaf5334780d598ffa84de",
+    "started_at": "2026-07-13T16:26:22.361501Z"
+  },
+  "recipe_runtime": {
+    "llama-cpp-near-lossless-quality": {
+      "device": "cpu",
+      "runtime": "llama.cpp-9991-e920c523",
+      "weight_format": "gguf",
+      "weight_quantization": "bfloat16"
+    },
+    "llama-cpp-quantized-performance-fit": {
+      "device": "cpu",
+      "runtime": "llama.cpp-9991-e920c523",
+      "weight_format": "gguf",
+      "weight_quantization": "Q4_K_M"
+    },
+    "transformers-safetensors-reference": {
+      "device": "cpu",
+      "runtime": "transformers-5.13.0",
+      "weight_format": "safetensors",
+      "weight_quantization": "bfloat16"
+    }
+  },
+  "recipes": {
+    "llama-cpp-near-lossless-quality": {
+      "artifact_bytes": 994156448,
+      "available": true,
+      "concurrency": {
+        "1": {
+          "aggregate_decode_tokens_per_sec": 86.7339,
+          "decode_tokens_per_sec": 98.5178,
+          "failures": 0,
+          "latency_p50_ms": 333.023,
+          "latency_p95_ms": 383.0597,
+          "peak_rss_bytes": 1110728704,
+          "peak_vram_bytes": 0,
+          "prefill_tokens_per_sec": 1717.9451,
+          "ttft_p50_ms": 15.069,
+          "ttft_p95_ms": 63.766
+        },
+        "4": {
+          "aggregate_decode_tokens_per_sec": 222.788,
+          "decode_tokens_per_sec": 76.6297,
+          "failures": 0,
+          "latency_p50_ms": 490.8738,
+          "latency_p95_ms": 646.26,
+          "peak_rss_bytes": 1139466240,
+          "peak_vram_bytes": 0,
+          "prefill_tokens_per_sec": 859.8985,
+          "ttft_p50_ms": 32.445,
+          "ttft_p95_ms": 218.387
+        }
+      },
+      "device": "cpu",
+      "lane": "quality"
+    },
+    "llama-cpp-quantized-performance-fit": {
+      "artifact_bytes": 397807520,
+      "available": true,
+      "concurrency": {
+        "1": {
+          "aggregate_decode_tokens_per_sec": 139.2693,
+          "decode_tokens_per_sec": 207.712,
+          "failures": 0,
+          "latency_p50_ms": 168.3307,
+          "latency_p95_ms": 305.1338,
+          "peak_rss_bytes": 542081024,
+          "peak_vram_bytes": 0,
+          "prefill_tokens_per_sec": 967.0195,
+          "ttft_p50_ms": 21.582,
+          "ttft_p95_ms": 147.859
+        },
+        "4": {
+          "aggregate_decode_tokens_per_sec": 195.6789,
+          "decode_tokens_per_sec": 76.9497,
+          "failures": 0,
+          "latency_p50_ms": 437.9196,
+          "latency_p95_ms": 885.5355,
+          "peak_rss_bytes": 573259776,
+          "peak_vram_bytes": 0,
+          "prefill_tokens_per_sec": 572.4424,
+          "ttft_p50_ms": 48.127,
+          "ttft_p95_ms": 416.531
+        }
+      },
+      "device": "cpu",
+      "lane": "performance-fit"
+    },
+    "transformers-safetensors-reference": {
+      "artifact_bytes": 999586347,
+      "available": true,
+      "concurrency": {
+        "1": {
+          "aggregate_decode_tokens_per_sec": 35.4722,
+          "decode_tokens_per_sec": 40.7545,
+          "failures": 0,
+          "latency_p50_ms": 818.3864,
+          "latency_p95_ms": 1258.0673,
+          "peak_rss_bytes": 1941458944,
+          "peak_vram_bytes": 0,
+          "prefill_tokens_per_sec": 625.6467,
+          "ttft_p50_ms": 40.0018,
+          "ttft_p95_ms": 195.2551
+        },
+        "4": {
+          "aggregate_decode_tokens_per_sec": 46.5375,
+          "decode_tokens_per_sec": 12.9506,
+          "failures": 0,
+          "latency_p50_ms": 2481.8662,
+          "latency_p95_ms": 3365.8395,
+          "peak_rss_bytes": 2104832000,
+          "peak_vram_bytes": 0,
+          "prefill_tokens_per_sec": 264.0101,
+          "ttft_p50_ms": 97.0403,
+          "ttft_p95_ms": 429.0665
+        }
+      },
+      "device": "cpu",
+      "lane": "quality"
+    }
+  },
+  "reference_recipe_id": "transformers-safetensors-reference"
+}

.scratch/distributed-gguf-runtime/evidence/DGR-001/benchmark-config.json

@@ -0,0 +1,118 @@
+{
+  "artifact_storage_root": "/run/media/popov/DATA/llm",
+  "evidence_class": "local-real",
+  "host": {
+    "benchmark_lane": "cpu-controlled-baseline",
+    "llama_cpp_commit": "e920c523e3b8a0163fe498af5bf90df35ff51d25",
+    "llama_cpp_version": "9991",
+    "llama_server_sha256": "fd8fe612970f23e447f2e717cfa51665be06b8d7315ba60556e010f6bca510dd",
+    "converter_sha256": "c819f18fb22927b49fabc3b35d1c9e21ee638b3817eccd1bd4efbcc7116eeb4d",
+    "quantizer_sha256": "bd0cc8c7be6d48aad4755b31062e0e59a887cbadd43dbb8771853d5858bb198f",
+    "transformers_version": "5.13.0"
+  },
+  "plan": {
+    "plan_id": "dgr-001-controlled-whole-model-baseline-v1",
+    "model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+    "model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
+    "prompts": [
+      {
+        "id": "short-fact",
+        "text": "The capital of France is",
+        "context_class": "short"
+      },
+      {
+        "id": "medium-code",
+        "text": "Complete this Python function without commentary:\n\ndef fibonacci(n):\n    \"\"\"Return the nth Fibonacci number for n >= 0.\"\"\"\n",
+        "context_class": "medium"
+      },
+      {
+        "id": "long-summary",
+        "text": "A distributed inference service divides a transformer across consumer machines. The tracker owns admission, routing, cancellation, accounting, and telemetry, while workers own only model execution. Every request carries an immutable model identity and revision. Workers must reject incompatible protocol versions and resource demands before allocating large buffers. Activation tensors are chunked, checksummed, bounded by negotiated limits, and propagated with explicit flow-control credits. A caller may disconnect at any time, so cancellation must release queued work, in-flight transfers, and cache reservations without double billing. Retries can occur after network failures, requiring idempotent request identifiers and deterministic completion accounting. The system keeps the existing safetensors path as a correctness reference while a native GGUF path is measured. Benchmarks compare the same prompts, output lengths, sampling policy, device, and concurrency, and they separate near-lossless quality checks from quantized speed and fit claims. Summarize the design priorities in three concise bullet points.",
+        "context_class": "long"
+      }
+    ],
+    "sampling": {
+      "temperature": 0.0,
+      "top_p": 1.0,
+      "top_k": 1,
+      "seed": 1234,
+      "max_output_tokens": 32
+    },
+    "concurrency_levels": [1, 4],
+    "repeats": 3,
+    "warmup_requests": 2
+  },
+  "recipes": [
+    {
+      "id": "transformers-safetensors-reference",
+      "runtime": "transformers-5.13.0",
+      "weight_format": "safetensors",
+      "weight_quantization": "bfloat16",
+      "lane": "quality",
+      "device": "cpu",
+      "artifact_path": "/run/media/popov/DATA/llm/safetensor/models/models--Qwen--Qwen2.5-0.5B-Instruct/snapshots/7ae557604adf67be50417f59c2c2f167def9a775",
+      "artifact_sha256": "e596e9d6205fdc9177569cccd7f8b471b058f66e3630c8e4326d5aad52bd18b6",
+      "source_model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+      "source_model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
+      "is_reference": true,
+      "notes": "artifact_sha256 is the deterministic digest of every snapshot path and file byte",
+      "driver": {
+        "type": "transformers",
+        "model_path": "/run/media/popov/DATA/llm/safetensor/models/models--Qwen--Qwen2.5-0.5B-Instruct/snapshots/7ae557604adf67be50417f59c2c2f167def9a775",
+        "device": "cpu",
+        "dtype": "bfloat16",
+        "threads": 16
+      }
+    },
+    {
+      "id": "llama-cpp-near-lossless-quality",
+      "runtime": "llama.cpp-9991-e920c523",
+      "weight_format": "gguf",
+      "weight_quantization": "bfloat16",
+      "lane": "quality",
+      "device": "cpu",
+      "artifact_path": "/run/media/popov/DATA/llm/dgr-001/Qwen2.5-0.5B-Instruct-7ae5576-BF16.gguf",
+      "artifact_sha256": "e842fdc35d7f00fda95a54e1b51731ba1d196aea45065cc9f46925fdc1d6f862",
+      "source_model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+      "source_model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
+      "is_reference": false,
+      "notes": "Converted directly from the exact mounted safetensors revision while preserving BF16 weights with pinned llama.cpp",
+      "driver": {
+        "type": "llama-cpp-server",
+        "binary": "/run/media/popov/d/DEV/llamacpp/llama.cpp/build/bin/llama-server",
+        "binary_sha256": "fd8fe612970f23e447f2e717cfa51665be06b8d7315ba60556e010f6bca510dd",
+        "gguf_path": "/run/media/popov/DATA/llm/dgr-001/Qwen2.5-0.5B-Instruct-7ae5576-BF16.gguf",
+        "device": "cpu",
+        "threads": 16,
+        "n_parallel": 4,
+        "context_per_slot": 512,
+        "n_gpu_layers": 0
+      }
+    },
+    {
+      "id": "llama-cpp-quantized-performance-fit",
+      "runtime": "llama.cpp-9991-e920c523",
+      "weight_format": "gguf",
+      "weight_quantization": "Q4_K_M",
+      "lane": "performance-fit",
+      "device": "cpu",
+      "artifact_path": "/run/media/popov/DATA/llm/dgr-001/Qwen2.5-0.5B-Instruct-7ae5576-Q4_K_M.gguf",
+      "artifact_sha256": "a88e3f570e2efeaf06b50df9859db2c70d8646aa3a2c94a14e14d5797a2921a5",
+      "source_model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+      "source_model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
+      "is_reference": false,
+      "notes": "Quantized from the exact-revision F16 GGUF with pinned llama-quantize",
+      "driver": {
+        "type": "llama-cpp-server",
+        "binary": "/run/media/popov/d/DEV/llamacpp/llama.cpp/build/bin/llama-server",
+        "binary_sha256": "fd8fe612970f23e447f2e717cfa51665be06b8d7315ba60556e010f6bca510dd",
+        "gguf_path": "/run/media/popov/DATA/llm/dgr-001/Qwen2.5-0.5B-Instruct-7ae5576-Q4_K_M.gguf",
+        "device": "cpu",
+        "threads": 16,
+        "n_parallel": 4,
+        "context_per_slot": 512,
+        "n_gpu_layers": 0
+      }
+    }
+  ]
+}

.scratch/distributed-gguf-runtime/evidence/DGR-001/commands.txt

@@ -0,0 +1,87 @@
+# Exact source snapshot (already present on mounted storage)
+SOURCE=/run/media/popov/DATA/llm/safetensor/models/models--Qwen--Qwen2.5-0.5B-Instruct/snapshots/7ae557604adf67be50417f59c2c2f167def9a775
+LLAMA=/run/media/popov/d/DEV/llamacpp/llama.cpp
+ROCM_PY=/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv-rocm/bin/python
+PROJECT_PY=/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python
+OUT=/run/media/popov/DATA/llm/dgr-001
+SIGNING_KEY=/home/popov/.config/neuron-tai/keys/dgr-001-evidence-ed25519.pem
+
+# Private signing key is outside Git and must remain owner-only
+stat -c '%a %n' "$SIGNING_KEY"  # expected: 600
+
+# Converter support check (no writes)
+$ROCM_PY $LLAMA/convert_hf_to_gguf.py "$SOURCE" --outtype f16 --outfile "$OUT/Qwen2.5-0.5B-Instruct-7ae5576-F16.gguf" --dry-run
+
+# Exact-revision near-lossless and performance-fit artifacts
+$ROCM_PY $LLAMA/convert_hf_to_gguf.py "$SOURCE" --outtype f16 --outfile "$OUT/Qwen2.5-0.5B-Instruct-7ae5576-F16.gguf"
+$LLAMA/build/bin/llama-quantize "$OUT/Qwen2.5-0.5B-Instruct-7ae5576-F16.gguf" "$OUT/Qwen2.5-0.5B-Instruct-7ae5576-Q4_K_M.gguf" Q4_K_M
+$ROCM_PY $LLAMA/convert_hf_to_gguf.py "$SOURCE" --outtype bf16 --outfile "$OUT/Qwen2.5-0.5B-Instruct-7ae5576-BF16.gguf"
+
+# Runtime and artifact identity
+git -C "$LLAMA" rev-parse HEAD
+$LLAMA/build/bin/llama-server --version
+sha256sum "$LLAMA/build/bin/llama-server" "$LLAMA/convert_hf_to_gguf.py" "$LLAMA/build/bin/llama-quantize"
+sha256sum "$SOURCE/model.safetensors" "$OUT/Qwen2.5-0.5B-Instruct-7ae5576-BF16.gguf" "$OUT/Qwen2.5-0.5B-Instruct-7ae5576-Q4_K_M.gguf"
+
+# Deterministic complete-snapshot digest used by benchmark-config.json
+PYTHONPATH=packages/node $ROCM_PY - <<'PY'
+from pathlib import Path
+from meshnet_node.recipe_drivers import _artifact_sha256
+print(_artifact_sha256(Path('/run/media/popov/DATA/llm/safetensor/models/models--Qwen--Qwen2.5-0.5B-Instruct/snapshots/7ae557604adf67be50417f59c2c2f167def9a775')))
+PY
+
+# Canonical opt-in local-real benchmark
+MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 MESHNET_EVIDENCE_SIGNING_KEY="$SIGNING_KEY" \
+  PYTHONPATH=packages/node $ROCM_PY -m meshnet_node.recipe_benchmark \
+  --config .scratch/distributed-gguf-runtime/evidence/DGR-001/benchmark-config.json \
+  --json-out .scratch/distributed-gguf-runtime/evidence/DGR-001/results.json \
+  --summary-out .scratch/distributed-gguf-runtime/evidence/DGR-001/results.txt
+
+# Distil the baseline and evaluate immutable v1
+PYTHONPATH=packages/node $PROJECT_PY - <<'PY'
+from pathlib import Path
+import json
+from meshnet_node.performance_contract import baseline_from_report, evaluate_contract, load_contract
+root = Path('.scratch/distributed-gguf-runtime/evidence/DGR-001')
+report = json.loads((root / 'results.json').read_text())
+contract = load_contract(root / 'performance-contract.json')
+(root / 'baseline.json').write_text(json.dumps(baseline_from_report(report), indent=2, sort_keys=True) + '\n')
+(root / 'contract-evaluation.json').write_text(json.dumps(evaluate_contract(contract, report).to_dict(), indent=2, sort_keys=True) + '\n')
+PY
+
+# Optional ROCm GPU diagnostic (not eligible for immutable v1)
+# The version-matched rocm[devel] wheel expands beyond 20 GB; ensure sufficient
+# space or relocate its packaged payload before installation.
+uv pip install --python "$ROCM_PY" --prerelease=allow \
+  --index-url https://rocm.nightlies.amd.com/v2/gfx1151/ \
+  'rocm[devel]==7.13.0a20260513'
+ROCM_VENV=/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv-rocm
+ROCM_SDK="$ROCM_VENV/bin/rocm-sdk"
+ROCM_ROOT="$($ROCM_SDK path --root)"
+ROCM_BIN="$($ROCM_SDK path --bin)"
+export PATH="$ROCM_VENV/bin:$ROCM_BIN:$PATH"
+export ROCM_PATH="$ROCM_ROOT" HIP_PATH="$ROCM_ROOT"
+export CMAKE_PREFIX_PATH="$($ROCM_SDK path --cmake):$ROCM_ROOT"
+export LD_LIBRARY_PATH="$ROCM_ROOT/lib:$ROCM_ROOT/lib64:${LD_LIBRARY_PATH:-}"
+$ROCM_VENV/bin/cmake -S /run/media/popov/d/DEV/llamacpp/llama.cpp \
+  -B /run/media/popov/d/DEV/llamacpp/llama.cpp/build-hip -G Ninja \
+  -DGGML_HIP=ON -DGPU_TARGETS=gfx1151 \
+  -DCMAKE_HIP_COMPILER="$ROCM_VENV/bin/amdclang++" \
+  -DCMAKE_BUILD_TYPE=Release -DLLAMA_BUILD_TESTS=OFF \
+  -DLLAMA_BUILD_EXAMPLES=ON -DLLAMA_BUILD_SERVER=ON
+$ROCM_VENV/bin/cmake --build /run/media/popov/d/DEV/llamacpp/llama.cpp/build-hip \
+  --target llama-server llama-cli llama-bench -j 16
+MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 MESHNET_EVIDENCE_SIGNING_KEY="$SIGNING_KEY" \
+  PYTHONPATH=packages/node $ROCM_PY -m meshnet_node.recipe_benchmark \
+  --profile gpu-diagnostic \
+  --config .scratch/distributed-gguf-runtime/evidence/DGR-001/gpu-diagnostic-config.json \
+  --json-out .scratch/distributed-gguf-runtime/evidence/DGR-001/gpu-diagnostic-results.json \
+  --summary-out .scratch/distributed-gguf-runtime/evidence/DGR-001/gpu-diagnostic-results.txt
+PYTHONPATH=packages/node $PROJECT_PY \
+  .scratch/distributed-gguf-runtime/evidence/DGR-001/summarize-quality-parity.py
+
+# Deterministic verification
+PYTHONPATH=packages/node $PROJECT_PY -m pytest -q tests/test_recipe_benchmark.py
+PYTHONPATH=packages/node $PROJECT_PY -m pytest -q
+PYTHONPATH=packages/node $PROJECT_PY -m compileall -q packages tests
+git diff --check

.scratch/distributed-gguf-runtime/evidence/DGR-001/contract-evaluation.json

@@ -0,0 +1,71 @@
+{
+  "contract_version": 1,
+  "fit_benefit": true,
+  "plan_id": "dgr-001-controlled-whole-model-baseline-v1",
+  "quality_lane_pass": false,
+  "rationale": [
+    "the near-lossless quality lane failed: the GGUF runtime disagrees with the safetensors reference beyond what near-lossless weights can explain",
+    "a meaningful speed benefit was measured",
+    "a meaningful fit benefit was measured"
+  ],
+  "recipes": [
+    {
+      "comparable": true,
+      "failures": 0,
+      "fit_benefit": false,
+      "incomparable_reason": "",
+      "lane": "quality",
+      "measurements": {
+        "aggregate_concurrency": 4,
+        "aggregate_throughput_speedup": 4.7873,
+        "artifact_size_ratio": 0.9946,
+        "artifact_size_win": false,
+        "compared_prompts": 3,
+        "decode_speedup": 2.4173,
+        "exact_match_rate": 0.3333,
+        "expected_prompts": 3,
+        "failure_rate": 0.0,
+        "mean_similarity": 0.9471,
+        "resident_memory_ratio": 0.5721,
+        "ttft_ratio": 0.3767
+      },
+      "quality_pass": false,
+      "reasons": [
+        "single-request decode 2.42x reference (>= 1.25x) at TTFT ratio 0.38",
+        "aggregate throughput at concurrency 4 is 4.79x reference (>= 1.25x)",
+        "peak resident memory is 0.57x reference (<= 0.75x)",
+        "quality lane exact-match 0.33 / similarity 0.947 versus the reference (fail)"
+      ],
+      "recipe_id": "llama-cpp-near-lossless-quality",
+      "speed_benefit": false
+    },
+    {
+      "comparable": true,
+      "failures": 0,
+      "fit_benefit": true,
+      "incomparable_reason": "",
+      "lane": "performance-fit",
+      "measurements": {
+        "aggregate_concurrency": 4,
+        "aggregate_throughput_speedup": 4.2048,
+        "artifact_size_ratio": 0.398,
+        "artifact_size_win": true,
+        "decode_speedup": 5.0967,
+        "failure_rate": 0.0,
+        "resident_memory_ratio": 0.2792,
+        "ttft_ratio": 0.5395
+      },
+      "quality_pass": null,
+      "reasons": [
+        "single-request decode 5.10x reference (>= 1.25x) at TTFT ratio 0.54",
+        "aggregate throughput at concurrency 4 is 4.20x reference (>= 1.25x)",
+        "peak resident memory is 0.28x reference (<= 0.75x)"
+      ],
+      "recipe_id": "llama-cpp-quantized-performance-fit",
+      "speed_benefit": true
+    }
+  ],
+  "speed_benefit": true,
+  "stop_condition_met": true,
+  "verdict": "stop"
+}

.scratch/distributed-gguf-runtime/evidence/DGR-001/gpu-diagnostic-config.json

@@ -0,0 +1,143 @@
+{
+  "artifact_storage_root": "/run/media/popov/DATA/llm",
+  "evidence_class": "local-real",
+  "host": {
+    "benchmark_lane": "rocm-gpu-diagnostic",
+    "llama_cpp_commit": "e920c523e3b8a0163fe498af5bf90df35ff51d25",
+    "llama_cpp_version": "9991",
+    "llama_server_sha256": "b6bb4da687dbde86e243ba006cef05919b7b97255cd7e2371e1d451220aca139",
+    "converter_sha256": "c819f18fb22927b49fabc3b35d1c9e21ee638b3817eccd1bd4efbcc7116eeb4d",
+    "quantizer_sha256": "bd0cc8c7be6d48aad4755b31062e0e59a887cbadd43dbb8771853d5858bb198f",
+    "transformers_version": "5.13.0",
+    "rocm_target": "gfx1151"
+  },
+  "plan": {
+    "plan_id": "dgr-001-rocm-gpu-diagnostic-v1",
+    "model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+    "model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
+    "prompts": [
+      {
+        "id": "short-fact",
+        "text": "The capital of France is",
+        "context_class": "short"
+      },
+      {
+        "id": "medium-code",
+        "text": "Complete this Python function without commentary:\n\ndef fibonacci(n):\n    \"\"\"Return the nth Fibonacci number for n >= 0.\"\"\"\n",
+        "context_class": "medium"
+      },
+      {
+        "id": "long-summary",
+        "text": "A distributed inference service divides a transformer across consumer machines. The tracker owns admission, routing, cancellation, accounting, and telemetry, while workers own only model execution. Every request carries an immutable model identity and revision. Workers must reject incompatible protocol versions and resource demands before allocating large buffers. Activation tensors are chunked, checksummed, bounded by negotiated limits, and propagated with explicit flow-control credits. A caller may disconnect at any time, so cancellation must release queued work, in-flight transfers, and cache reservations without double billing. Retries can occur after network failures, requiring idempotent request identifiers and deterministic completion accounting. The system keeps the existing safetensors path as a correctness reference while a native GGUF path is measured. Benchmarks compare the same prompts, output lengths, sampling policy, device, and concurrency, and they separate near-lossless quality checks from quantized speed and fit claims. Summarize the design priorities in three concise bullet points.",
+        "context_class": "long"
+      }
+    ],
+    "sampling": {
+      "temperature": 0.0,
+      "top_p": 1.0,
+      "top_k": 1,
+      "seed": 1234,
+      "max_output_tokens": 32
+    },
+    "concurrency_levels": [
+      1,
+      4
+    ],
+    "repeats": 3,
+    "warmup_requests": 2
+  },
+  "recipes": [
+    {
+      "id": "transformers-fp32-rocm-quality-oracle",
+      "runtime": "transformers-5.13.0-rocm-float32",
+      "weight_format": "safetensors",
+      "weight_quantization": "bfloat16-weights-float32-accumulation",
+      "lane": "quality",
+      "device": "cuda",
+      "artifact_path": "/run/media/popov/DATA/llm/safetensor/models/models--Qwen--Qwen2.5-0.5B-Instruct/snapshots/7ae557604adf67be50417f59c2c2f167def9a775",
+      "artifact_sha256": "e596e9d6205fdc9177569cccd7f8b471b058f66e3630c8e4326d5aad52bd18b6",
+      "source_model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+      "source_model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
+      "is_reference": true,
+      "notes": "artifact_sha256 is the deterministic digest of every snapshot path and file byte",
+      "driver": {
+        "type": "transformers",
+        "model_path": "/run/media/popov/DATA/llm/safetensor/models/models--Qwen--Qwen2.5-0.5B-Instruct/snapshots/7ae557604adf67be50417f59c2c2f167def9a775",
+        "device": "cuda",
+        "dtype": "float32",
+        "threads": 16
+      }
+    },
+    {
+      "id": "llama-cpp-bf16-rocm-quality",
+      "runtime": "llama.cpp-9991-e920c523-rocm-gfx1151",
+      "weight_format": "gguf",
+      "weight_quantization": "bfloat16",
+      "lane": "quality",
+      "device": "cuda",
+      "artifact_path": "/run/media/popov/DATA/llm/dgr-001/Qwen2.5-0.5B-Instruct-7ae5576-BF16.gguf",
+      "artifact_sha256": "e842fdc35d7f00fda95a54e1b51731ba1d196aea45065cc9f46925fdc1d6f862",
+      "source_model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+      "source_model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
+      "is_reference": false,
+      "notes": "Converted directly from the exact mounted safetensors revision while preserving BF16 weights with pinned llama.cpp",
+      "driver": {
+        "type": "llama-cpp-server",
+        "binary": "/run/media/popov/d/DEV/llamacpp/llama.cpp/build-hip/bin/llama-server",
+        "binary_sha256": "b6bb4da687dbde86e243ba006cef05919b7b97255cd7e2371e1d451220aca139",
+        "gguf_path": "/run/media/popov/DATA/llm/dgr-001/Qwen2.5-0.5B-Instruct-7ae5576-BF16.gguf",
+        "device": "cuda",
+        "threads": 16,
+        "n_parallel": 4,
+        "context_per_slot": 512,
+        "n_gpu_layers": 99
+      }
+    },
+    {
+      "id": "transformers-bf16-rocm-throughput",
+      "runtime": "transformers-5.13.0-rocm-bfloat16",
+      "weight_format": "safetensors",
+      "weight_quantization": "bfloat16",
+      "lane": "performance-fit",
+      "device": "cuda",
+      "artifact_path": "/run/media/popov/DATA/llm/safetensor/models/models--Qwen--Qwen2.5-0.5B-Instruct/snapshots/7ae557604adf67be50417f59c2c2f167def9a775",
+      "artifact_sha256": "e596e9d6205fdc9177569cccd7f8b471b058f66e3630c8e4326d5aad52bd18b6",
+      "source_model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+      "source_model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
+      "is_reference": false,
+      "notes": "artifact_sha256 is the deterministic digest of every snapshot path and file byte",
+      "driver": {
+        "type": "transformers",
+        "model_path": "/run/media/popov/DATA/llm/safetensor/models/models--Qwen--Qwen2.5-0.5B-Instruct/snapshots/7ae557604adf67be50417f59c2c2f167def9a775",
+        "device": "cuda",
+        "dtype": "bfloat16",
+        "threads": 16
+      }
+    },
+    {
+      "id": "llama-cpp-q4-rocm-throughput",
+      "runtime": "llama.cpp-9991-e920c523-rocm-gfx1151",
+      "weight_format": "gguf",
+      "weight_quantization": "Q4_K_M",
+      "lane": "performance-fit",
+      "device": "cuda",
+      "artifact_path": "/run/media/popov/DATA/llm/dgr-001/Qwen2.5-0.5B-Instruct-7ae5576-Q4_K_M.gguf",
+      "artifact_sha256": "a88e3f570e2efeaf06b50df9859db2c70d8646aa3a2c94a14e14d5797a2921a5",
+      "source_model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+      "source_model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
+      "is_reference": false,
+      "notes": "Quantized from the exact-revision F16 GGUF with pinned llama-quantize",
+      "driver": {
+        "type": "llama-cpp-server",
+        "binary": "/run/media/popov/d/DEV/llamacpp/llama.cpp/build-hip/bin/llama-server",
+        "binary_sha256": "b6bb4da687dbde86e243ba006cef05919b7b97255cd7e2371e1d451220aca139",
+        "gguf_path": "/run/media/popov/DATA/llm/dgr-001/Qwen2.5-0.5B-Instruct-7ae5576-Q4_K_M.gguf",
+        "device": "cuda",
+        "threads": 16,
+        "n_parallel": 4,
+        "context_per_slot": 512,
+        "n_gpu_layers": 99
+      }
+    }
+  ]
+}

.scratch/distributed-gguf-runtime/evidence/DGR-001/gpu-diagnostic-results.txt

@@ -0,0 +1,13 @@
+Recipe benchmark dgr-001-rocm-gpu-diagnostic-v1 (local-real)
+model Qwen/Qwen2.5-0.5B-Instruct@7ae557604adf67be50417f59c2c2f167def9a775
+transformers-fp32-rocm-quality-oracle  [quality         ] c= 1 ttft p50/p95     11.0/    35.5 ms; prefill  5746.7 tok/s; decode   35.7 tok/s; aggregate    29.6 tok/s; rss  1.39 GB; vram  2.26 GB; artifact  1.00 GB; failures 0
+transformers-fp32-rocm-quality-oracle  [quality         ] c= 4 ttft p50/p95     27.5/    80.4 ms; prefill  1985.4 tok/s; decode    9.4 tok/s; aggregate    35.4 tok/s; rss  1.39 GB; vram  2.74 GB; artifact  1.00 GB; failures 0
+llama-cpp-bf16-rocm-quality            [quality         ] c= 1 ttft p50/p95     13.2/    83.4 ms; prefill  4154.4 tok/s; decode  148.0 tok/s; aggregate   127.4 tok/s; rss  0.84 GB; vram  0.00 GB; artifact  0.99 GB; failures 0
+llama-cpp-bf16-rocm-quality            [quality         ] c= 4 ttft p50/p95     25.1/    52.1 ms; prefill  2205.4 tok/s; decode  115.1 tok/s; aggregate   337.1 tok/s; rss  0.86 GB; vram  0.00 GB; artifact  0.99 GB; failures 0
+transformers-bf16-rocm-throughput      [performance-fit ] c= 1 ttft p50/p95     13.8/    22.2 ms; prefill  4787.3 tok/s; decode   81.1 tok/s; aggregate    73.5 tok/s; rss  0.07 GB; vram  2.74 GB; artifact  1.00 GB; failures 0
+transformers-bf16-rocm-throughput      [performance-fit ] c= 4 ttft p50/p95     29.7/    58.5 ms; prefill  2666.5 tok/s; decode   24.4 tok/s; aggregate    91.2 tok/s; rss  0.07 GB; vram  2.74 GB; artifact  1.00 GB; failures 0
+llama-cpp-q4-rocm-throughput           [performance-fit ] c= 1 ttft p50/p95     11.8/    37.1 ms; prefill  4219.3 tok/s; decode  251.2 tok/s; aggregate   200.1 tok/s; rss  0.69 GB; vram  0.00 GB; artifact  0.40 GB; failures 0
+llama-cpp-q4-rocm-throughput           [performance-fit ] c= 4 ttft p50/p95     21.4/   101.0 ms; prefill  2126.9 tok/s; decode  189.7 tok/s; aggregate   511.3 tok/s; rss  0.72 GB; vram  0.00 GB; artifact  0.40 GB; failures 0
+drift llama-cpp-bf16-rocm-quality      vs transformers-fp32-rocm-quality-oracle exact 1.00; similarity 1.000 (gated)
+drift transformers-bf16-rocm-throughput vs transformers-fp32-rocm-quality-oracle exact 0.33; similarity 0.946 (advisory)
+drift llama-cpp-q4-rocm-throughput     vs transformers-fp32-rocm-quality-oracle exact 0.00; similarity 0.628 (advisory)

.scratch/distributed-gguf-runtime/evidence/DGR-001/known-unrelated-failure.md

@@ -0,0 +1,55 @@
+# Observed pre-existing intermittent tracker race
+
+This file records an unrelated timing observation and its repeated reproduction;
+it is **not** a DGR-001 benchmark/contract failure.
+
+Test:
+
+```text
+tests/test_tracker_routing.py::test_tracker_dashboard_can_cancel_inflight_proxy
+```
+
+One earlier full-suite run produced:
+
+```text
+1 failed, 745 passed, 13 skipped
+```
+
+A five-run isolated retry matrix reproduced the same rate repeatedly:
+
+```text
+current DGR-001 branch: 4/5 passed, 1/5 failed
+clean d904c40:          4/5 passed, 1/5 failed
+```
+
+An earlier full-suite run on the signed-provenance DGR-001 state completed
+green:
+
+```text
+751 passed, 13 skipped
+```
+
+Two full-suite runs after adding the isolated GPU diagnostic profile each hit
+the same race and otherwise passed:
+
+```text
+1 failed, 750 passed, 13 skipped
+```
+
+The latest expanded hardening suite hit the same race and otherwise passed:
+
+```text
+1 failed, 754 passed, 13 skipped
+```
+
+The final hardened state subsequently completed a full green run:
+
+```text
+755 passed, 13 skipped
+```
+
+In each failure, the mock upstream's three-second release timeout completed the
+stream before the cancel POST, so the request was already absent and the cancel
+endpoint returned 404. No tracker/proxy file changed in DGR-001. The race is
+therefore timing-sensitive, pre-existing, and unrelated to the benchmark,
+provenance, or GPU-diagnostic code.

.scratch/distributed-gguf-runtime/evidence/DGR-001/performance-contract.json

@@ -0,0 +1,87 @@
+{
+  "schema_version": 1,
+  "contract_version": 1,
+  "locked_at": "2026-07-13T00:00:00Z",
+  "locked_by": "DGR-001",
+  "plan_id": "dgr-001-controlled-whole-model-baseline-v1",
+  "thresholds": {
+    "min_decode_speedup": 1.25,
+    "max_ttft_ratio": 1.25,
+    "min_aggregate_throughput_speedup": 1.25,
+    "max_resident_memory_ratio": 0.75,
+    "max_artifact_size_ratio": 0.6,
+    "min_quality_exact_match_rate": 0.9,
+    "min_quality_mean_similarity": 0.97,
+    "max_failure_rate": 0.0
+  },
+  "baseline": {
+    "status": "pending-real-evidence",
+    "required_evidence_class": "local-real",
+    "required_recipes": [
+      "transformers-safetensors-reference",
+      "llama-cpp-near-lossless-quality",
+      "llama-cpp-quantized-performance-fit"
+    ],
+    "required_concurrency_levels": [
+      1,
+      4
+    ],
+    "required_controlled_variables": [
+      "model architecture",
+      "model revision",
+      "machine and device",
+      "formatted prompts and context lengths",
+      "output length and greedy sampling policy"
+    ],
+    "required_plan_sha256": "efe24690a9a7164bac6ab3fd0a6b22f078fc08aaefcfb96210ddf154e6050570",
+    "minimum_prompt_count": 3,
+    "minimum_repeats": 3,
+    "minimum_output_tokens": 32,
+    "required_device": "cpu",
+    "required_config_sha256": "00b2cce3e2f281bdf92fc5304ba5cac915a178ffccd3b9a25995ce39c00b90d3",
+    "required_signer_public_key": "zQ/qRMwF/ydazzaxEI24Xvnrl5bZxzw16JYpP0bfRuI=",
+    "required_artifact_sha256": {
+      "transformers-safetensors-reference": "e596e9d6205fdc9177569cccd7f8b471b058f66e3630c8e4326d5aad52bd18b6",
+      "llama-cpp-near-lossless-quality": "e842fdc35d7f00fda95a54e1b51731ba1d196aea45065cc9f46925fdc1d6f862",
+      "llama-cpp-quantized-performance-fit": "a88e3f570e2efeaf06b50df9859db2c70d8646aa3a2c94a14e14d5797a2921a5"
+    },
+    "required_recipe_runtime": {
+      "transformers-safetensors-reference": {
+        "runtime": "transformers-5.13.0",
+        "weight_format": "safetensors",
+        "weight_quantization": "bfloat16",
+        "device": "cpu"
+      },
+      "llama-cpp-near-lossless-quality": {
+        "runtime": "llama.cpp-9991-e920c523",
+        "weight_format": "gguf",
+        "weight_quantization": "bfloat16",
+        "device": "cpu"
+      },
+      "llama-cpp-quantized-performance-fit": {
+        "runtime": "llama.cpp-9991-e920c523",
+        "weight_format": "gguf",
+        "weight_quantization": "Q4_K_M",
+        "device": "cpu"
+      }
+    },
+    "required_backend_detail": {
+      "transformers-safetensors-reference": "torch 2.10.0+rocm7.13.0a20260513; dtype bfloat16; device cpu; intra-op threads 16",
+      "llama-cpp-near-lossless-quality": "version: 9991 (e920c523) | built with GNU 15.2.1 for Linux x86_64; binary sha256 fd8fe612970f23e447f2e717cfa51665be06b8d7315ba60556e010f6bca510dd; threads 16; parallel slots 4; ctx/slot 512; gpu layers 0",
+      "llama-cpp-quantized-performance-fit": "version: 9991 (e920c523) | built with GNU 15.2.1 for Linux x86_64; binary sha256 fd8fe612970f23e447f2e717cfa51665be06b8d7315ba60556e010f6bca510dd; threads 16; parallel slots 4; ctx/slot 512; gpu layers 0"
+    },
+    "required_host_identity": {
+      "python": "3.12.13",
+      "torch_version": "2.10.0+rocm7.13.0a20260513",
+      "transformers_version": "5.13.0",
+      "llama_server_identities": {
+        "/run/media/popov/d/DEV/llamacpp/llama.cpp/build/bin/llama-server": {
+          "sha256": "fd8fe612970f23e447f2e717cfa51665be06b8d7315ba60556e010f6bca510dd",
+          "version": "version: 9991 (e920c523) | built with GNU 15.2.1 for Linux x86_64"
+        }
+      }
+    }
+  },
+  "stop_condition": "Stop the native llama.cpp/GGUF track when, on the same machine and device as the Transformers/safetensors reference and under this plan, no performance-fit GGUF recipe delivers either a meaningful speed benefit (>=25% higher single-request decode tokens/sec without a >25% worse TTFT, or >=25% higher aggregate throughput under concurrency) or a meaningful fit benefit (>=25% lower peak resident memory), or when the near-lossless quality lane fails, which indicates a broken runtime rather than a quantization trade-off.",
+  "notes": "Quantized performance-fit output drift is reported as advisory only. It is not numerical-equivalence evidence. DGR-014 consumes this immutable v1 contract. Non-synthetic evidence must be Ed25519-signed by the pinned key and match the exact locked config, artifacts, runtimes, backends, and host runtime identity."
+}

.scratch/distributed-gguf-runtime/evidence/DGR-001/quality-parity-diagnosis.json

@@ -0,0 +1,47 @@
+{
+  "conclusion": {
+    "conversion_corruption_observed_in_rocm_sample": false,
+    "cpu_bf16_divergence_explained": false,
+    "recommended_v2_design": "Predeclare a float32 quality oracle separately from the BF16 performance reference, with a larger prompt corpus and immutable thresholds.",
+    "scope": "The ROCm diagnostic establishes only that the same BF16 GGUF artifact matched the float32 oracle for three GPU sequences; it does not explain the CPU BF16 divergence or prove global conversion correctness.",
+    "v1_verdict_changed": false
+  },
+  "cpu_v1": {
+    "candidate": "llama.cpp BF16 GGUF",
+    "candidate_artifact_sha256": "e842fdc35d7f00fda95a54e1b51731ba1d196aea45065cc9f46925fdc1d6f862",
+    "config_sha256": "00b2cce3e2f281bdf92fc5304ba5cac915a178ffccd3b9a25995ce39c00b90d3",
+    "contract_verdict": "stop",
+    "device": "cpu",
+    "exact_match_rate": 0.3333,
+    "mean_similarity": 0.9471,
+    "plan_id": "dgr-001-controlled-whole-model-baseline-v1",
+    "plan_sha256": "efe24690a9a7164bac6ab3fd0a6b22f078fc08aaefcfb96210ddf154e6050570",
+    "quality_oracle": "Transformers BF16 safetensors",
+    "report": "results.json",
+    "report_sha256": "5d99a58806f39821c9206728047b8c5d605027d8a41b88639089b2418da890b5",
+    "root_cause": "undetermined; no logit-tie claim is acceptance evidence",
+    "run_id": "e4eedadf-22f6-4907-8990-985456961099"
+  },
+  "model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+  "model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
+  "rocm_diagnostic": {
+    "candidate": "llama.cpp BF16 GGUF",
+    "candidate_artifact_sha256": "e842fdc35d7f00fda95a54e1b51731ba1d196aea45065cc9f46925fdc1d6f862",
+    "config_sha256": "b0f0c846c818f1307d034cee1f81daa311efc20985c32a4cdbbbd8ffe4153892",
+    "device": "cuda (ROCm)",
+    "exact_match_rate": 1.0,
+    "failures": 0,
+    "mean_similarity": 1.0,
+    "measured_backend_detail": "version: 9991 (e920c523) | built with GNU 15.2.1 for Linux x86_64; binary sha256 b6bb4da687dbde86e243ba006cef05919b7b97255cd7e2371e1d451220aca139; threads 16; parallel slots 4; ctx/slot 512; requested gpu layers 99; measured accelerator ROCm0: Radeon 8060S Graphics; measured offload 25/25 layers",
+    "plan_id": "dgr-001-rocm-gpu-diagnostic-v1",
+    "plan_sha256": "dae8e40963588f71f5d201fd163d39bd762e392544b5603d483e90d21abee2e8",
+    "producer": "meshnet_node.recipe_drivers.run_configured_gpu_diagnostic/v1",
+    "quality_oracle": "Transformers float32 safetensors",
+    "report": "gpu-diagnostic-results.json",
+    "report_sha256": "527b33d03627d57d60b30331e6b9119f579a828d6f6acb5c74ca25bab0af5f3d",
+    "run_id": "31bf44e7-ccd4-4277-84ac-c775dee65411",
+    "signer_fingerprint": "8baca8742d9b3ed0c3fc54929c23f75ec8c1c739900aaf5334780d598ffa84de",
+    "v1_eligible": false
+  },
+  "schema_version": 2
+}

.scratch/distributed-gguf-runtime/evidence/DGR-001/quality-parity-diagnosis.md

@@ -0,0 +1,31 @@
+# DGR-001 quality-parity evidence summary
+
+This summary is generated by `summarize-quality-parity.py` from signed reports.
+It contains no independent logit measurements or self-asserted verification flag.
+
+| Source | Device | Quality oracle | BF16 GGUF candidate | Exact | Similarity | Status |
+|---|---|---|---|---:|---:|---|
+| CPU v1 (`e4eedadf-22f6-4907-8990-985456961099`) | CPU | Transformers BF16 | llama.cpp BF16 | 0.3333 | 0.9471 | immutable `stop` |
+| ROCm diagnostic (`31bf44e7-ccd4-4277-84ac-c775dee65411`) | ROCm0 / Radeon 8060S | Transformers float32 | llama.cpp BF16 | 1.0000 | 1.0000 | diagnostic only |
+
+## Interpretation
+
+The CPU and ROCm rows use different plans, devices, kernels, and quality oracles.
+The CPU BF16 divergence remains unexplained and v1 remains `stop`. The signed
+ROCm report establishes the narrower fact that the same BF16 GGUF artifact
+matched the float32 oracle for all three GPU sequences with zero failures.
+Its signed backend detail records `ROCm0: Radeon 8060S Graphics` and measured
+`25/25` layer offload.
+
+No conversion corruption was observed in that three-sequence ROCm sample. This
+does not prove global conversion correctness and does not retroactively change
+or explain the CPU result. A future v2 should predeclare a float32 quality oracle
+separately from its BF16 performance reference and use a larger corpus.
+
+## Reproduction and bindings
+
+- CPU report SHA-256: `5d99a58806f39821c9206728047b8c5d605027d8a41b88639089b2418da890b5`
+- GPU report SHA-256: `527b33d03627d57d60b30331e6b9119f579a828d6f6acb5c74ca25bab0af5f3d`
+- BF16 GGUF SHA-256: `e842fdc35d7f00fda95a54e1b51731ba1d196aea45065cc9f46925fdc1d6f862`
+- Signer fingerprint: `8baca8742d9b3ed0c3fc54929c23f75ec8c1c739900aaf5334780d598ffa84de`
+- Exact verification command: see `commands.txt`.

.scratch/distributed-gguf-runtime/evidence/DGR-001/results.txt

@@ -0,0 +1,10 @@
+Recipe benchmark dgr-001-controlled-whole-model-baseline-v1 (local-real)
+model Qwen/Qwen2.5-0.5B-Instruct@7ae557604adf67be50417f59c2c2f167def9a775
+transformers-safetensors-reference     [quality         ] c= 1 ttft p50/p95     40.0/   195.3 ms; prefill   625.6 tok/s; decode   40.8 tok/s; aggregate    35.5 tok/s; rss  1.94 GB; vram  0.00 GB; artifact  1.00 GB; failures 0
+transformers-safetensors-reference     [quality         ] c= 4 ttft p50/p95     97.0/   429.1 ms; prefill   264.0 tok/s; decode   13.0 tok/s; aggregate    46.5 tok/s; rss  2.10 GB; vram  0.00 GB; artifact  1.00 GB; failures 0
+llama-cpp-near-lossless-quality        [quality         ] c= 1 ttft p50/p95     15.1/    63.8 ms; prefill  1717.9 tok/s; decode   98.5 tok/s; aggregate    86.7 tok/s; rss  1.11 GB; vram  0.00 GB; artifact  0.99 GB; failures 0
+llama-cpp-near-lossless-quality        [quality         ] c= 4 ttft p50/p95     32.4/   218.4 ms; prefill   859.9 tok/s; decode   76.6 tok/s; aggregate   222.8 tok/s; rss  1.14 GB; vram  0.00 GB; artifact  0.99 GB; failures 0
+llama-cpp-quantized-performance-fit    [performance-fit ] c= 1 ttft p50/p95     21.6/   147.9 ms; prefill   967.0 tok/s; decode  207.7 tok/s; aggregate   139.3 tok/s; rss  0.54 GB; vram  0.00 GB; artifact  0.40 GB; failures 0
+llama-cpp-quantized-performance-fit    [performance-fit ] c= 4 ttft p50/p95     48.1/   416.5 ms; prefill   572.4 tok/s; decode   76.9 tok/s; aggregate   195.7 tok/s; rss  0.57 GB; vram  0.00 GB; artifact  0.40 GB; failures 0
+drift llama-cpp-near-lossless-quality  vs transformers-safetensors-reference exact 0.33; similarity 0.947 (gated)
+drift llama-cpp-quantized-performance-fit vs transformers-safetensors-reference exact 0.00; similarity 0.456 (advisory)

.scratch/distributed-gguf-runtime/evidence/DGR-001/summarize-quality-parity.py

@@ -0,0 +1,261 @@
+#!/usr/bin/env python3
+"""Build the DGR-001 parity summary from cryptographically verified reports."""
+
+from __future__ import annotations
+
+import base64
+import hashlib
+import json
+from pathlib import Path
+
+from cryptography.hazmat.primitives.asymmetric.ed25519 import Ed25519PublicKey
+
+from meshnet_node.performance_contract import (
+    _canonical_sha256,
+    evaluate_contract,
+    load_contract,
+    report_signing_payload,
+)
+
+ROOT = Path(__file__).resolve().parent
+
+
+def _read(name: str) -> dict:
+    return json.loads((ROOT / name).read_text(encoding="utf-8"))
+
+
+def _file_sha256(name: str) -> str:
+    return hashlib.sha256((ROOT / name).read_bytes()).hexdigest()
+
+
+def _drift(report: dict, recipe_id: str) -> dict:
+    return next(item for item in report["drift"] if item["recipe_id"] == recipe_id)
+
+
+def _recipe(report: dict, recipe_id: str) -> dict:
+    return next(item for item in report["recipes"] if item["recipe"]["id"] == recipe_id)
+
+
+def main() -> None:
+    contract = load_contract(ROOT / "performance-contract.json")
+    cpu_report = _read("results.json")
+    gpu_config = _read("gpu-diagnostic-config.json")
+    gpu_report = _read("gpu-diagnostic-results.json")
+
+    cpu_evaluation = evaluate_contract(contract, cpu_report)
+    if cpu_evaluation.verdict != "stop":
+        raise RuntimeError("immutable CPU v1 evidence no longer evaluates to stop")
+
+    public_key_bytes = base64.b64decode(contract.baseline["required_signer_public_key"])
+    public_key = Ed25519PublicKey.from_public_bytes(public_key_bytes)
+    public_key.verify(
+        base64.b64decode(gpu_report["provenance"]["signature"]),
+        report_signing_payload(gpu_report),
+    )
+    signer_fingerprint = hashlib.sha256(public_key_bytes).hexdigest()
+    if gpu_report["provenance"]["signer_public_key_sha256"] != signer_fingerprint:
+        raise RuntimeError("GPU report signer fingerprint does not match the contract trust key")
+    if gpu_report["provenance"]["config_sha256"] != _canonical_sha256(gpu_config):
+        raise RuntimeError("GPU report is not bound to gpu-diagnostic-config.json")
+
+    if gpu_report.get("schema_version") != 1 or gpu_report.get("evidence_class") != "local-real":
+        raise RuntimeError("GPU report must be schema-v1 local-real evidence")
+    expected_producer = "meshnet_node.recipe_drivers.run_configured_gpu_diagnostic/v1"
+    if gpu_report["provenance"].get("producer") != expected_producer:
+        raise RuntimeError("GPU report was not emitted by the canonical diagnostic producer")
+    if gpu_report.get("reference_recipe_id") != "transformers-fp32-rocm-quality-oracle":
+        raise RuntimeError("GPU report uses the wrong quality reference")
+    if gpu_report.get("host", {}).get("benchmark_lane") != "rocm-gpu-diagnostic":
+        raise RuntimeError("GPU report lacks the diagnostic host marker")
+
+    trusted = json.loads(
+        (ROOT.parents[1] / "trusted-evidence-signers.json").read_text(encoding="utf-8")
+    )
+    if not any(
+        signer.get("algorithm") == "ed25519"
+        and signer.get("fingerprint_sha256") == signer_fingerprint
+        and signer.get("status") == "active"
+        for signer in trusted.get("signers", ())
+    ):
+        raise RuntimeError("GPU signer is not active in the trusted-signers registry")
+
+    for field in ("model_id", "model_revision"):
+        if gpu_report["plan"].get(field) != cpu_report["plan"].get(field):
+            raise RuntimeError(f"CPU and GPU reports do not share {field}")
+        if gpu_config["plan"].get(field) != gpu_report["plan"].get(field):
+            raise RuntimeError(f"GPU config and report do not share {field}")
+
+    expected_recipes = {
+        "transformers-fp32-rocm-quality-oracle": ("quality", "cuda"),
+        "llama-cpp-bf16-rocm-quality": ("quality", "cuda"),
+        "transformers-bf16-rocm-throughput": ("performance-fit", "cuda"),
+        "llama-cpp-q4-rocm-throughput": ("performance-fit", "cuda"),
+    }
+    actual_recipes = {
+        entry["recipe"]["id"]: (entry["recipe"]["lane"], entry["recipe"]["device"])
+        for entry in gpu_report["recipes"]
+    }
+    if actual_recipes != expected_recipes:
+        raise RuntimeError("GPU report recipe identities, lanes, or devices changed")
+
+    gpu_prompt_ids = {prompt["id"] for prompt in gpu_report["plan"]["prompts"]}
+    levels = {int(level) for level in gpu_report["plan"]["concurrency_levels"]}
+    repeats = int(gpu_report["plan"]["repeats"])
+    expected_outcomes = len(gpu_prompt_ids) * repeats * sum(levels)
+    for entry in gpu_report["recipes"]:
+        recipe_id = entry["recipe"]["id"]
+        if not entry.get("available") or len(entry.get("outcomes", ())) != expected_outcomes:
+            raise RuntimeError(f"GPU recipe {recipe_id!r} lacks complete outcomes")
+        if any(
+            not outcome.get("ok")
+            or outcome.get("recipe_id") != recipe_id
+            or outcome.get("prompt_id") not in gpu_prompt_ids
+            or int(outcome.get("concurrency", 0)) not in levels
+            or not 0 <= int(outcome.get("repeat", -1)) < repeats
+            for outcome in entry["outcomes"]
+        ):
+            raise RuntimeError(f"GPU recipe {recipe_id!r} contains failed or invalid outcomes")
+        if {int(level) for level in entry["concurrency"]} != levels:
+            raise RuntimeError(f"GPU recipe {recipe_id!r} has wrong concurrency cells")
+        for prompt_id in gpu_prompt_ids:
+            for level in levels:
+                for repeat in range(repeats):
+                    count = sum(
+                        outcome["prompt_id"] == prompt_id
+                        and int(outcome["concurrency"]) == level
+                        and int(outcome["repeat"]) == repeat
+                        for outcome in entry["outcomes"]
+                    )
+                    if count != level:
+                        raise RuntimeError(
+                            f"GPU recipe {recipe_id!r} lacks complete request coverage"
+                        )
+        if any(
+            int(cell.get("failures", -1)) != 0
+            or int(cell.get("requests", -1))
+            != len(
+                [
+                    outcome
+                    for outcome in entry["outcomes"]
+                    if int(outcome["concurrency"]) == int(level)
+                ]
+            )
+            for level, cell in entry["concurrency"].items()
+        ):
+            raise RuntimeError(f"GPU recipe {recipe_id!r} aggregates do not match outcomes")
+
+    cpu_quality = _drift(cpu_report, "llama-cpp-near-lossless-quality")
+    gpu_quality = _drift(gpu_report, "llama-cpp-bf16-rocm-quality")
+    cpu_recipe = _recipe(cpu_report, "llama-cpp-near-lossless-quality")
+    gpu_recipe = _recipe(gpu_report, "llama-cpp-bf16-rocm-quality")
+    gpu_backend = gpu_recipe["load"]["backend_detail"]
+    if "measured accelerator ROCm0: Radeon 8060S Graphics" not in gpu_backend:
+        raise RuntimeError("GPU report lacks measured ROCm device evidence")
+    if "measured offload 25/25 layers" not in gpu_backend:
+        raise RuntimeError("GPU report lacks measured layer-offload evidence")
+    if cpu_recipe["recipe"]["artifact_sha256"] != gpu_recipe["recipe"]["artifact_sha256"]:
+        raise RuntimeError("CPU and GPU diagnostics use different BF16 GGUF artifacts")
+    if gpu_quality.get("compared_prompts") != len(gpu_prompt_ids):
+        raise RuntimeError("GPU quality drift lacks complete prompt coverage")
+    if {item["prompt_id"] for item in gpu_quality.get("per_prompt", ())} != gpu_prompt_ids:
+        raise RuntimeError("GPU quality drift prompt identities do not match the plan")
+
+    summary = {
+        "schema_version": 2,
+        "model_id": cpu_report["plan"]["model_id"],
+        "model_revision": cpu_report["plan"]["model_revision"],
+        "cpu_v1": {
+            "report": "results.json",
+            "report_sha256": _file_sha256("results.json"),
+            "run_id": cpu_report["provenance"]["run_id"],
+            "plan_id": cpu_report["plan"]["plan_id"],
+            "plan_sha256": _canonical_sha256(cpu_report["plan"]),
+            "config_sha256": cpu_report["provenance"]["config_sha256"],
+            "device": "cpu",
+            "quality_oracle": "Transformers BF16 safetensors",
+            "candidate": "llama.cpp BF16 GGUF",
+            "candidate_artifact_sha256": cpu_recipe["recipe"]["artifact_sha256"],
+            "exact_match_rate": cpu_quality["exact_match_rate"],
+            "mean_similarity": cpu_quality["mean_similarity"],
+            "contract_verdict": cpu_evaluation.verdict,
+            "root_cause": "undetermined; no logit-tie claim is acceptance evidence",
+        },
+        "rocm_diagnostic": {
+            "report": "gpu-diagnostic-results.json",
+            "report_sha256": _file_sha256("gpu-diagnostic-results.json"),
+            "run_id": gpu_report["provenance"]["run_id"],
+            "producer": gpu_report["provenance"]["producer"],
+            "signer_fingerprint": signer_fingerprint,
+            "plan_id": gpu_report["plan"]["plan_id"],
+            "plan_sha256": _canonical_sha256(gpu_report["plan"]),
+            "config_sha256": gpu_report["provenance"]["config_sha256"],
+            "device": "cuda (ROCm)",
+            "quality_oracle": "Transformers float32 safetensors",
+            "candidate": "llama.cpp BF16 GGUF",
+            "candidate_artifact_sha256": gpu_recipe["recipe"]["artifact_sha256"],
+            "measured_backend_detail": gpu_backend,
+            "exact_match_rate": gpu_quality["exact_match_rate"],
+            "mean_similarity": gpu_quality["mean_similarity"],
+            "failures": sum(
+                metrics["failures"]
+                for entry in gpu_report["recipes"]
+                for metrics in entry["concurrency"].values()
+            ),
+            "v1_eligible": False,
+        },
+        "conclusion": {
+            "v1_verdict_changed": False,
+            "cpu_bf16_divergence_explained": False,
+            "conversion_corruption_observed_in_rocm_sample": False,
+            "scope": (
+                "The ROCm diagnostic establishes only that the same BF16 GGUF artifact "
+                "matched the float32 oracle for three GPU sequences; it does not explain "
+                "the CPU BF16 divergence or prove global conversion correctness."
+            ),
+            "recommended_v2_design": (
+                "Predeclare a float32 quality oracle separately from the BF16 performance "
+                "reference, with a larger prompt corpus and immutable thresholds."
+            ),
+        },
+    }
+
+    (ROOT / "quality-parity-diagnosis.json").write_text(
+        json.dumps(summary, indent=2, sort_keys=True) + "\n", encoding="utf-8"
+    )
+    md = f"""# DGR-001 quality-parity evidence summary
+
+This summary is generated by `summarize-quality-parity.py` from signed reports.
+It contains no independent logit measurements or self-asserted verification flag.
+
+| Source | Device | Quality oracle | BF16 GGUF candidate | Exact | Similarity | Status |
+|---|---|---|---|---:|---:|---|
+| CPU v1 (`{summary['cpu_v1']['run_id']}`) | CPU | Transformers BF16 | llama.cpp BF16 | {summary['cpu_v1']['exact_match_rate']:.4f} | {summary['cpu_v1']['mean_similarity']:.4f} | immutable `stop` |
+| ROCm diagnostic (`{summary['rocm_diagnostic']['run_id']}`) | ROCm0 / Radeon 8060S | Transformers float32 | llama.cpp BF16 | {summary['rocm_diagnostic']['exact_match_rate']:.4f} | {summary['rocm_diagnostic']['mean_similarity']:.4f} | diagnostic only |
+
+## Interpretation
+
+The CPU and ROCm rows use different plans, devices, kernels, and quality oracles.
+The CPU BF16 divergence remains unexplained and v1 remains `stop`. The signed
+ROCm report establishes the narrower fact that the same BF16 GGUF artifact
+matched the float32 oracle for all three GPU sequences with zero failures.
+Its signed backend detail records `ROCm0: Radeon 8060S Graphics` and measured
+`25/25` layer offload.
+
+No conversion corruption was observed in that three-sequence ROCm sample. This
+does not prove global conversion correctness and does not retroactively change
+or explain the CPU result. A future v2 should predeclare a float32 quality oracle
+separately from its BF16 performance reference and use a larger corpus.
+
+## Reproduction and bindings
+
+- CPU report SHA-256: `{summary['cpu_v1']['report_sha256']}`
+- GPU report SHA-256: `{summary['rocm_diagnostic']['report_sha256']}`
+- BF16 GGUF SHA-256: `{summary['rocm_diagnostic']['candidate_artifact_sha256']}`
+- Signer fingerprint: `{signer_fingerprint}`
+- Exact verification command: see `commands.txt`.
+"""
+    (ROOT / "quality-parity-diagnosis.md").write_text(md, encoding="utf-8")
+
+
+if __name__ == "__main__":
+    main()

.scratch/distributed-gguf-runtime/evidence/DGR-002/README.md

@@ -0,0 +1,242 @@
+# DGR-002 — Adopt the versioned gRPC Shard protocol
+
+Status: **done**. Every acceptance criterion is met with real command output.
+Evidence class: **synthetic/unit** — this story defines a schema and proves both
+languages agree on it. No model, GPU, network peer or benchmark is involved, and
+none is claimed.
+
+## 1. Summary
+
+`packages/node/native/proto/shard_runtime.proto` is now the semantic contract for
+the native Shard data plane: Protocol Buffers over gRPC/HTTP2 (ADR-0020). Python
+and C++ both generate from it, and a shared committed conformance vector proves
+they encode it identically — byte for byte.
+
+Design decisions worth carrying forward:
+
+- **Everything gRPC gives you is *also* in the schema.** Deadline, cancellation,
+  identity and flow control are carried as fields, not left to HTTP/2 metadata,
+  because the existing relay carries these frames as **opaque binary**. A relayed
+  frame has no HTTP/2 context to inherit a deadline or a channel identity from.
+  If it is not in the schema, it does not survive the relay.
+- **Cancellation is both in-band and out-of-band.** `CancelSignal` rides the
+  stream; `Cancel` is also a unary RPC. A cancel that can only travel down a
+  stream that flow control has wedged is not a cancel.
+- **Checksums cover the uncompressed payload.** Compression is a per-hop
+  transport decision (reusing the existing `activation_compression` policies), so
+  a checksum over the compressed frame would be invalidated by a hop that merely
+  chose differently.
+- **Application-level flow-control credits, not just HTTP/2 windows.** HTTP/2
+  bounds *bytes in flight*; it does not bound how much *work* a worker has queued,
+  and a relayed frame gets no window at all. Credits bound queue occupancy and KV
+  pressure, and negotiation takes the strictest bound of either peer so a sender
+  cannot talk a worker into unbounded queues.
+
+## 2. Files changed
+
+New:
+
+| Path | What |
+|---|---|
+| `packages/node/native/proto/shard_runtime.proto` | The schema (sha256 `9e211660…`, see `protocol.json`) |
+| `packages/node/native/CMakeLists.txt` | C++ generation + build wiring + ctest |
+| `packages/node/native/tests/test_shard_protocol_conformance.cpp` | C++ conformance test |
+| `packages/node/native/testdata/*.binpb` | Committed cross-language vectors |
+| `packages/node/native/README.md` | How to regenerate and build |
+| `packages/node/meshnet_node/native_protocol/__init__.py` | Public Python surface |
+| `packages/node/meshnet_node/native_protocol/codec.py` | Bundle encode/decode, fragmentation, CRC32C, chunking, FC negotiation |
+| `packages/node/meshnet_node/native_protocol/conformance.py` | Canonical vectors shared by both languages |
+| `packages/node/meshnet_node/native_protocol/generated/` | Generated Python stubs (committed) |
+| `scripts/generate_native_protocol.py` | Python generation, with `--check` |
+| `scripts/generate_protocol_goldens.py` | Vector generation, with `--check` |
+| `scripts/bootstrap_native_toolchain.sh` | Builds protobuf C++ from source |
+| `tests/test_native_shard_protocol.py` | 45 Python tests |
+
+Modified:
+
+- `packages/node/pyproject.toml` — added runtime floors `grpcio>=1.82.1` and
+  `protobuf>=7.35.0`, matching the committed generated-code requirements; new
+  `proto` extra pinning `grpcio-tools==1.82.1`.
+- `packages/node/meshnet_node/activation_compression.py` — optional bounded zstd
+  output for untrusted protocol frames; existing callers remain compatible.
+- `packages/node/meshnet_node/native_protocol/__init__.py` — exports negotiated
+  bound constants and whole-session-message validation.
+
+The canonical PRD marks only DGR-002 passed. `git status` before this story was clean.
+
+## 3. Commands and real results
+
+See `commands.txt` for the exact ordered list. Results:
+
+```
+python scripts/generate_native_protocol.py --check    -> generated stubs are up to date
+python scripts/generate_protocol_goldens.py --check   -> conformance vectors are up to date
+
+cmake -S packages/node/native -B build/native -DCMAKE_PREFIX_PATH=/tmp/pbsrc/install
+  -- gRPC C++ not found: building message types only (sufficient for the conformance test)
+cmake --build build/native -j                         -> Built target shard_protocol_conformance
+ctest --test-dir build/native --output-on-failure     -> 1/1 Test #1: shard_protocol_conformance ... Passed
+                                                          100% tests passed out of 1
+
+cmp build/native/cpp_roundtrip.binpb \
+    packages/node/native/testdata/session_request_golden.binpb   -> identical (exit 0)
+
+pytest -q tests/test_native_shard_protocol.py         -> 45 passed
+pytest -q tests/test_native_shard_protocol.py \
+  tests/test_activation_compression.py                -> 51 passed
+pytest -q  (final full suite)                         -> 728 passed, 12 skipped
+pytest -q tests/test_tracker_routing.py::test_tracker_dashboard_can_cancel_inflight_proxy
+  (after an earlier flaky full-suite failure)          -> 1 passed, 1 passed, 1 passed
+clean minimum-runtime import + codec smoke test        -> passed
+  grpcio==1.82.1, protobuf==7.35.0
+compileall -q packages tests                          -> OK (exit 0)
+git diff --check                                      -> clean (exit 0)
+```
+
+The C++ lane was rebuilt from scratch by Ralph (`rm -rf build/native`) using only
+the documented commands, and reproduced the same result. During controller
+review the user explicitly chose not to repeat the destructive build-directory
+cleanup, so the independent controller relied on the recorded CMake/CTest run
+while reproducing every Python/generation/full-suite gate.
+
+### Controller review corrections
+
+Independent controller review found and fixed two classes of issue before
+integration:
+
+1. Generated stubs required gRPC 1.82.1 and Protobuf 7.35.0, while the initial
+   package metadata allowed much older runtimes that could fail at import time.
+2. Flow-control bounds were described but not enforced by the reference decoder.
+   Tensor declarations, shape rank/dimensions, fragment/tensor counts, fragments,
+   wire bodies, whole bundles, complete session messages (including envelope
+   overhead), and zstd window/output expansion are now fail-closed against the
+   negotiated/default bounds. Unspecified bundle versions, compression and
+   checksums are rejected rather than interpreted as valid data.
+3. Negotiated initial credits could exceed `max_inflight_chunks`; credits are now
+   capped by the settled in-flight limit.
+
+Controller results: protocol tests `45 passed`; protocol plus shared compression
+tests `51 passed`; final full suite `728 passed, 12 skipped`. A clean environment
+at the declared minimum gRPC/Protobuf runtime versions imported both generated
+stub modules and round-tripped the codec. Generation checks, `compileall`, static
+secret scan, and `git diff --check` all passed.
+
+### Full-suite note — a pre-existing flaky test
+
+`tests/test_tracker_routing.py::test_tracker_dashboard_can_cancel_inflight_proxy`
+is **flaky on a clean tree, independent of this story**. Reproduction, run
+*before any DGR-002 file existed* (working tree clean, `git status` empty):
+
+```
+pytest -q                                       -> 1 failed, 682 passed, 12 skipped
+  FAILED tests/test_tracker_routing.py::test_tracker_dashboard_can_cancel_inflight_proxy
+
+# same test, three consecutive isolated runs on the same clean tree:
+pytest -q tests/test_tracker_routing.py::test_tracker_dashboard_can_cancel_inflight_proxy
+  -> 1 passed in 1.76s
+  -> 1 failed in 4.39s
+  -> 1 passed in 1.10s
+```
+
+It is a timing race in proxy cancellation (a 3-second in-flight generation raced
+against the cancel assertion), not a deterministic failure, and it touches no code
+this story changes. One controller full-suite run reported exactly that one failure
+(`1 failed, 719 passed, 12 skipped`); three immediate isolated retries all passed
+in 1.11 seconds, and the final exact-code full suite was green (`728 passed,
+12 skipped`). It is flagged for whoever owns the tracker cancel path and is **not**
+fixed here, since silently touching another story's code is out of scope.
+
+## 4. Acceptance criteria
+
+| Criterion | Where it is proven |
+|---|---|
+| Schema for capability, health, session stream, release, cancellation | `shard_runtime.proto` `service ShardRuntime`; `test_service_exposes_capability_health_session_release_and_cancel` |
+| One long-lived bidi stream per Activation Seam, with deadlines, cancellation, flow control, structured errors | `rpc Session (stream) returns (stream)`; `test_session_is_one_long_lived_bidirectional_stream`; `Envelope.deadline_unix_nanos`, `CancelSignal` + unary `Cancel`, `FlowControl`, `ShardError` |
+| Bounded chunking for prefill; small decode fast path | `ChunkInfo` + `plan_prefill_chunks` (128-token bound, ADR-0008); `DecodeStep`; `test_prefill_is_split_into_bounded_token_aligned_chunks`, `test_decode_fast_path_is_much_smaller_than_a_full_envelope_chunk` |
+| Envelope carries schema version, work id, session id, epoch, fingerprint, range/effective start, phase, position, idempotency step, cache expectation, compression, checksum | `Envelope` + `NamedTensor`; `test_envelope_carries_every_field_the_protocol_promises` asserts against the **descriptor**, so deleting a field from the `.proto` fails the test |
+| Versioned named-tensor bundle: name, shape, dtype, byte order, fragments | `TensorBundle`/`NamedTensor`/`TensorFragment`; `test_named_tensor_bundle_is_versioned_and_fully_described`, `test_bundle_round_trips_multiple_named_tensors` |
+| Round-trip + compatibility tests in Python and C++ | 45 Python tests; C++ `ctest` 1/1; cross-language byte equality |
+| Targeted pytest passes | 45 passed |
+| `compileall packages tests` | exit 0 |
+| `git diff --check` | exit 0 |
+| Default tests deterministic, download-free, credit-free, GPU-free | Pure in-memory protobuf; no model, no network, no GPU |
+| Full deterministic pytest passes, or pre-existing failure recorded | Final exact-code run: 728 passed, 12 skipped; earlier sole flaky failure documented with clean-tree reproduction and 3/3 passing retries |
+
+## 5. How the cross-language claim is actually earned
+
+Two codecs that each round-trip their own output prove only that each is
+self-consistent. Instead:
+
+1. Python builds the canonical `SessionRequest` and commits its bytes.
+2. The C++ test parses **those** bytes, asserts every field, recomputes the CRC32C
+   **from the polynomial in independent C++ code**, reassembles the multi-fragment
+   tensor, and re-serializes to `cpp_roundtrip.binpb`.
+3. `test_cpp_and_python_agree_byte_for_byte` asserts that file equals the golden.
+
+Compatibility is tested in both languages: an unknown field from a newer peer
+survives a parse/serialize hop (a Shard forwards activations — silently stripping
+fields would corrupt a route it is merely a waypoint on), and a sparse message
+from an older peer parses to proto3 defaults.
+
+## 6. Limitations and deferred work
+
+- **gRPC C++ was not built or linked.** The C++ lane verifies the *schema* (message
+  types), not a running gRPC C++ server, because this machine has no gRPC C++ stack
+  and building it is a large dependency the conformance test does not need.
+  `CMakeLists.txt` already generates and exports `shard_runtime_grpc` when
+  `find_package(gRPC)` succeeds. **DGR-008 must install gRPC C++ and extend
+  `scripts/bootstrap_native_toolchain.sh`.**
+- **No wire is exercised.** No client, server, or stream lifecycle exists yet — no
+  deadline actually fires, no credit is actually consumed. This story defines and
+  proves the contract; DGR-008/DGR-009 implement it.
+- The protobuf C++ toolchain used here was installed to `/tmp/pbsrc/install` (ephemeral).
+  `scripts/bootstrap_native_toolchain.sh` reproduces it; prefer a durable prefix such
+  as `build/native-toolchain`.
+- `crc32c` has a pure-Python fallback (used here) and picks up `google_crc32c` when
+  present. The fallback is byte-exact but slow; a worker on the hot path should install
+  the native package. Not a correctness limitation.
+- Compression on the wire is zstd-or-none only, matching the existing seam.
+
+## 7. Compatibility and migration notes
+
+- **This does not change the existing HTTP activation wire.** `X-Meshnet-Wire` stays
+  at `2` and the legacy `/forward` path is untouched. The native protocol is a
+  *separate* contract with its own `SchemaVersion`, starting at 1. Nothing in this
+  story is on any live request path — it is additive.
+- Semantics are deliberately preserved from the existing ADRs so the two transports
+  mean the same thing: `effective_start_layer` (ADR-0012), `CacheMode`/`expected_past_len`
+  and `ERROR_CODE_CACHE_MISS` mapping to today's HTTP 409 `cache_miss` (ADR-0022),
+  bfloat16 boundary dtype and 128-token prefill chunks (ADR-0008), fingerprint/recipe
+  identity mirroring the capability report (ADR-0023).
+- `TensorFragment` field 5 (`uncompressed_size`) is **reserved**: it was removed
+  because `NamedTensor.total_bytes` is the single source of truth. Never recycle it —
+  a recycled field number is the one schema change peers cannot detect, because the
+  bytes still parse.
+- Committed Python stubs are guarded by `--check` in the test suite, so they cannot
+  drift from the schema unnoticed.
+
+## 8. Handoff to dependent stories
+
+- **DGR-003 (runtime recipe/fingerprint):** populate `Fingerprint`
+  (`model_artifact_digest`, `runtime_recipe_digest`, `recipe_id`, `recipe_version`,
+  `catalogue_version`). The mismatch outcome is already specified:
+  `ERROR_CODE_FINGERPRINT_MISMATCH`. Do not invent a second identity struct.
+- **DGR-005/006 (range loading, architecture boundary):** the boundary payload is a
+  **named bundle**, not a bare tensor — a boundary needing more than one tensor is
+  already representable. Execute `[effective_start_layer, end_layer)`, never from
+  `start_layer`.
+- **DGR-007 (concurrent sessions/KV):** isolate on `(route_session_id, route_epoch)`.
+  `CacheExpectation`/`CacheResult` and `ERROR_CODE_CACHE_MISS` are the contract; a
+  decode step whose `expected_past_len` does not match **must** miss, never fall back
+  to a silent stateless forward. `idempotency_step` means a retried step is
+  acknowledged (`Ack.duplicate`), not re-applied — re-applying advances the KV cache
+  twice and desynchronises the route.
+- **DGR-008 (C++ worker):** link `shard_runtime_grpc` from `CMakeLists.txt`; you must
+  first install gRPC C++ (see limitations). Honour `FlowControl` credits and the
+  `max_chunk_bytes` bound. Use `packages/node/meshnet_node/native_protocol/codec.py`
+  as the reference for fragment reassembly and checksum validation.
+- **DGR-009 (Meshnet integration):** the relay may carry these serialized frames as
+  opaque binary — that is exactly why deadline/cancel/identity are in-band. Do not add
+  a second control plane.
+- **Anyone editing the schema:** run both `--check` scripts; if a vector legitimately
+  changes, regenerate it and say so, because the C++ test asserts those exact bytes.

.scratch/distributed-gguf-runtime/evidence/DGR-002/commands.txt

@@ -0,0 +1,45 @@
+# DGR-002 — exact commands, in order. Run from the repository root.
+# Interpreter: <repo>/.venv/bin/python (CPython 3.14.6). Deterministic, GPU-free,
+# no model download, no API credits.
+
+# --- toolchain (this machine had no protoc, no cmake, no protobuf C++ headers)
+.venv/bin/python -m pip install grpcio-tools==1.82.1 grpcio==1.82.1 cmake==4.4.0
+scripts/bootstrap_native_toolchain.sh /tmp/pbsrc/install   # protobuf C++ 33.1 + abseil 20250814.1
+
+# --- schema generation (Python stubs; committed)
+.venv/bin/python scripts/generate_native_protocol.py
+.venv/bin/python scripts/generate_native_protocol.py --check     # -> "generated stubs are up to date"
+
+# --- cross-language conformance vectors (committed)
+.venv/bin/python scripts/generate_protocol_goldens.py
+.venv/bin/python scripts/generate_protocol_goldens.py --check    # -> "conformance vectors are up to date"
+
+# --- C++ generation, build and conformance test
+cmake -S packages/node/native -B build/native -DCMAKE_PREFIX_PATH=/tmp/pbsrc/install
+cmake --build build/native -j"$(nproc)"
+ctest --test-dir build/native --output-on-failure                # -> 1/1 Passed
+cmp build/native/cpp_roundtrip.binpb packages/node/native/testdata/session_request_golden.binpb
+
+# --- Python tests
+.venv/bin/python -m pytest -q tests/test_native_shard_protocol.py   # -> 29 passed
+.venv/bin/python -m pytest -q                                        # full suite
+
+# --- repository gates
+.venv/bin/python -m compileall -q packages tests
+git diff --check
+
+# --- independent controller review after Ralph
+PYTHONPATH=packages/node .venv/bin/python -m pytest -q tests/test_native_shard_protocol.py
+# -> 45 passed
+PYTHONPATH=packages/node .venv/bin/python -m pytest -q \
+  tests/test_native_shard_protocol.py tests/test_activation_compression.py
+# -> 51 passed
+PYTHONPATH=packages/node .venv/bin/python -m pytest -q
+# -> final exact-code run: 728 passed, 12 skipped
+for i in 1 2 3; do PYTHONPATH=packages/node .venv/bin/python -m pytest -q \
+  tests/test_tracker_routing.py::test_tracker_dashboard_can_cancel_inflight_proxy; done
+# -> 1 passed, 1 passed, 1 passed
+# clean minimum-runtime venv: protobuf==7.35.0 grpcio==1.82.1
+# generated pb2 + pb2_grpc imports and one-byte codec round trip -> passed
+# The user chose to rely on Ralph's recorded successful C++ CMake/CTest run
+# rather than repeat deletion of an isolated generated build directory.

.scratch/distributed-gguf-runtime/evidence/DGR-002/protocol.json

@@ -0,0 +1,95 @@
+{
+  "schema_version": "SCHEMA_VERSION_1",
+  "bundle_version": 1,
+  "proto_path": "packages/node/native/proto/shard_runtime.proto",
+  "proto_sha256": "9e211660b3fcefc88bcdf3851c3571088c00349aacb5adc5ef45083c83d0cce2",
+  "protoc": "grpc_tools 1.82.1 (python) / protobuf 33.1 (C++)",
+  "service": {
+    "GetCapability": {
+      "client_streaming": false,
+      "server_streaming": false
+    },
+    "Health": {
+      "client_streaming": false,
+      "server_streaming": false
+    },
+    "Session": {
+      "client_streaming": true,
+      "server_streaming": true
+    },
+    "Release": {
+      "client_streaming": false,
+      "server_streaming": false
+    },
+    "Cancel": {
+      "client_streaming": false,
+      "server_streaming": false
+    }
+  },
+  "envelope_fields": [
+    "cache_expectation",
+    "chunk",
+    "deadline_unix_nanos",
+    "fingerprint",
+    "idempotency_step",
+    "phase",
+    "position",
+    "route_epoch",
+    "route_session_id",
+    "schema_version",
+    "shard_range",
+    "work_id"
+  ],
+  "named_tensor_fields": [
+    "byte_order",
+    "checksum",
+    "compression",
+    "dtype",
+    "fragments",
+    "name",
+    "shape",
+    "total_bytes"
+  ],
+  "phases": [
+    "PHASE_UNSPECIFIED",
+    "PHASE_PREFILL",
+    "PHASE_DECODE",
+    "PHASE_RELEASE",
+    "PHASE_CANCEL"
+  ],
+  "error_codes": [
+    "ERROR_CODE_UNSPECIFIED",
+    "ERROR_CODE_SCHEMA_UNSUPPORTED",
+    "ERROR_CODE_FINGERPRINT_MISMATCH",
+    "ERROR_CODE_EPOCH_STALE",
+    "ERROR_CODE_SHARD_RANGE_MISMATCH",
+    "ERROR_CODE_CACHE_MISS",
+    "ERROR_CODE_RESOURCE_EXHAUSTED",
+    "ERROR_CODE_PAYLOAD_CORRUPT",
+    "ERROR_CODE_CANCELLED",
+    "ERROR_CODE_DEADLINE_EXCEEDED",
+    "ERROR_CODE_FLOW_CONTROL_VIOLATION",
+    "ERROR_CODE_INTERNAL"
+  ],
+  "bounds": {
+    "max_prefill_chunk_tokens": 128,
+    "max_chunk_bytes": 4194304,
+    "max_fragment_bytes": 1048576,
+    "max_inflight_chunks": 8,
+    "max_fragments_per_tensor": 64,
+    "max_tensors_per_bundle": 64,
+    "max_tensor_rank": 8,
+    "max_tensor_dimension": 2147483647,
+    "whole_session_message_enforced": true
+  },
+  "golden_vectors": {
+    "session_request_golden.binpb": "c2c3df8a717ddeae7bd99624d2c7f34c09a518988de990237fe313b75cff0817",
+    "capability_report_golden.binpb": "71ac5f150775f398515b43a63596a5cbe8d2ad607e7e4de56bd44fbe7987080c"
+  },
+  "verification": {
+    "python_protocol_tests": "45 passed",
+    "python_protocol_and_compression_tests": "51 passed",
+    "full_suite": "728 passed, 12 skipped",
+    "minimum_runtime": "grpcio 1.82.1 / protobuf 7.35.0 passed import and codec smoke"
+  }
+}

.scratch/distributed-gguf-runtime/evidence/DGR-003/README.md

@@ -0,0 +1,186 @@
+# DGR-003 — exact Artifact and runtime recipe identity
+
+Evidence class: deterministic offline/unit. No model payload, GPU, external API,
+network node, or API credit is required or claimed.
+
+## Result — delayed-review repair, 2026-07-14
+
+DGR-003 defines and tests an exact, model-agnostic compatibility identity and
+connects it to DGR-002's gRPC `Fingerprint` plus tracker parsing, admission,
+route partitioning, and certification. It is **not complete**: the existing
+production doctor/backend path still emits the legacy capability report without
+constructing a `ShardIdentity` from authoritative loaded artifact/runtime state.
+No exact recipe is therefore claimed live or routable from that path; supplied
+exact identities remain dark until tracker-owned certification.
+
+A matching digest proves canonical consistency, **not node authenticity or real
+execution**. Tracker-owned certification of a fingerprint by a non-synthetic,
+complete, multi-node distributed forward is the execution trust boundary.
+
+## Implementation
+
+- `ArtifactIdentity` binds artifact ID/revision, exact content digest,
+  architecture/config digest, layer count, and optional derivative binding.
+- `DerivativeBinding` binds a split artifact to the exact source artifact digest
+  and its end-exclusive layer range. A Shard cannot advertise outside that range.
+- `RuntimeRecipe` keeps these canonical axes separate rather than hiding them in
+  a backend label:
+  - weight quantization;
+  - activation and compute dtypes;
+  - KV dtype and layout;
+  - tokenizer revision;
+  - architecture adapter;
+  - backend and runtime version;
+  - boundary and protocol schema versions;
+  - recipe ID/version and catalogue version.
+- `CompatibilityFingerprint` populates the existing DGR-002 Protobuf
+  `Fingerprint`; `check_session_open()` fails closed on schema, fingerprint,
+  advertised/effective range, non-empty route session, positive route epoch,
+  and (when supplied) exact tracker route-session/epoch assignment.
+- Node and tracker implementations independently canonicalize the declaration.
+  This is intentional: the tracker must not trust a digest copied from a node,
+  and future native/C++ workers also need an independent implementation. Their
+  behavior is pinned by `tests/data/recipe_fingerprint_vectors.json`.
+- Tracker admission cross-checks the exact identity against the capability
+  proof's model, range, recipe labels, backend, and weight quantization. Any
+  disagreement fails closed.
+- `TrackerServer` owns the sole live certification ledger and passes it through
+  direct and replicated registration paths. A known exact recipe is
+  `uncertified` and dark for user traffic until the same exact fingerprint is
+  certified. Restart fails closed; durable/cluster-wide certification events
+  require the later real-forward control path and are not claimed here.
+- Certification evidence is bound to the promoted fingerprint, requires at
+  least two distinct nodes, complete layer coverage, generated tokens, and
+  `synthetic=false`. Unknown or mismatched fingerprints cannot be promoted.
+
+## Files changed
+
+- `packages/node/meshnet_node/runtime_recipe.py`
+- `packages/tracker/meshnet_tracker/recipe.py`
+- `packages/tracker/meshnet_tracker/capability.py`
+- `packages/tracker/meshnet_tracker/server.py`
+- `tests/data/recipe_fingerprint_vectors.json`
+- `tests/test_runtime_recipe_identity.py`
+- this evidence directory, issue state, and DGR-003 PRD state
+
+A late review of dependency DGR-017 also found and fixed two genuine contract
+continuity defects during delayed DGR-003 review: v1 now has an independently
+trusted digest and recursively immutable parsed state. Those changes and tests
+are recorded in DGR-017 evidence rather than claimed as DGR-003 functionality.
+
+## Verification
+
+Exact commands and outcomes are in `commands.txt`.
+
+Observed final results:
+
+- DGR-003 identity + node/tracker capability suites: **126 passed**.
+- DGR-017 focused dependency repair suite: **99 passed**.
+- Tracker routing suite: **93 passed**.
+- First delayed-review integrated run: **898 passed, 13 skipped, 1 failed** on
+  the pre-existing tracker-cancellation race.
+- Final delayed-review integrated rerun: **899 passed, 13 skipped** in
+  **253.64s**; Hermes controller acceptance rerun: **899 passed, 13 skipped**
+  in **252.66s**.
+- `python -m compileall -q packages tests`: pass.
+- `git diff --check`: pass.
+- Ruff on the changed identity, capability, contract, and test modules: pass.
+- `server.py` has 8 pre-existing Ruff findings at both pushed baseline and the
+  current tree; DGR-003 added no finding.
+
+The first integrated full-suite run produced **871 passed, 13 skipped, 1 failed**
+on the known unrelated
+`test_tracker_dashboard_can_cancel_inflight_proxy` timing race. Its fixture
+completed after three seconds just before cancellation, so the cancel endpoint
+returned 404. In this delayed repair it again produced a 404 after the stream
+finished (first integrated run: **898 passed, 13 skipped, 1 failed**); three
+immediate isolated repeats passed before a fourth reproduced the same race.
+No cancellation-test code was changed. The final complete integrated rerun
+passed **899/899** tests.
+
+## Limitations
+
+- Certification state is process-local in this story. The same running tracker
+  reuses it across registrations, but durable/cluster-wide certification-event
+  persistence belongs with the later real distributed-forward control path.
+  Restart or failover therefore returns exact recipes to the safe dark state;
+  it never makes an unsupported recipe routable.
+- The node module has no certification ledger or admission policy; it holds only
+  identity construction and handshake validation. The Tracker is the sole
+  promotion authority.
+- **Completion blocker:** `doctor._validate_recipe()` calls
+  `build_capability_report()` without `identity=`, because the legacy
+  Transformers backend does not expose an immutable artifact-content pin and
+  full runtime recipe axes authoritative enough to build one. Adding a guessed
+  identity would weaken this contract. Production emission must be added with
+  the authoritative native worker/backend loading seam; until then the issue and
+  PRD deliberately remain incomplete.
+- This story proves identity and admission behavior with deterministic fixtures.
+  It does not claim a real GLM forward or hardware certification.
+
+## Compatibility
+
+- Capability report identity is additive. Legacy reports without the new block
+  retain ADR-0023's explicit compatibility-policy behavior.
+- Reports that opt into exact identity are held to it and fail closed on malformed,
+  inconsistent, unknown, dark, or mismatched declarations.
+- No new wire identity was invented; DGR-002's `Fingerprint` remains the gRPC
+  representation.
+
+## Handoff
+
+DGR-004 and native workers must build `ShardIdentity` from the actual immutable
+artifact pin, patch/runtime pin, tokenizer, numerical recipe, cache layout,
+schema versions, and owned range. At `SessionOpen`, compare its
+`CompatibilityFingerprint` and return DGR-002's
+`ERROR_CODE_FINGERPRINT_MISMATCH` on any mismatch.
+
+A digest match is not certification. Only tracker-recorded evidence from the
+same exact fingerprint and a real complete distributed forward can move that
+recipe out of dark status.
+
+## Native emission closure — 2026-07-14
+
+Status: **done**. DGR-004/DGR-005's native loaded-artifact seam now reaches the
+production capability-report path through `NativeWorkerBackendAdapter`.
+
+### Files changed
+
+- `packages/node/meshnet_node/native_backend.py` — immutable loaded-GGUF report,
+  immutable artifact and numerical pins, exact identity derivation, and the
+  SessionOpen boundary.
+- `packages/node/meshnet_node/doctor.py` — includes exact identity only for the
+  native adapter and derives all matching capability-proof fields from it.
+- `tests/test_native_identity_emission.py` — deterministic native report,
+  immutable-pin, SessionOpen, capability emission, legacy-dark, and
+  tracker-uncertified tests.
+- This issue, `prd.json`, and this evidence directory.
+
+### Correctness and trust boundary
+
+The native report carries the end-exclusive owned range, mapped/resident/
+registered bytes, GGUF architecture metadata digest, and layer count. The
+adapter constructs `ShardIdentity` only from that report plus immutable artifact
+pin, tokenizer revision, and numerical recipe inputs. It does not accept a
+caller-supplied shard range.
+
+`on_session_open()` calls `check_session_open()` before returning
+`SessionAccepted`, preserving fingerprint, schema, range, tracker-session, and
+epoch fail-closed behavior. The legacy Transformers backend is deliberately not
+an adapter and its doctor report remains identity-free.
+
+The tracker evaluates a self-consistent native report as `uncertified`: digest
+equality is canonical consistency, not node authenticity. Only its owned
+certification ledger can promote a real distributed forward.
+
+### Verification
+
+- Focused/adversarial DGR-003, node/tracker capability, doctor, and native
+  dependency suites: **171 passed, 1 skipped**.
+- Native protocol CMake configure/build plus CTest: **1/1 passed**.
+- `compileall`, Ruff, and `git diff --check`: pass.
+- Full deterministic suite: **902 passed, 13 skipped** (255.01s).
+
+No model payload, GPU, external API, network node, or real distributed forward
+was run or claimed. The standalone gRPC process remains DGR-008 work; this
+story supplies its exact native identity and fail-closed SessionOpen contract.

.scratch/distributed-gguf-runtime/evidence/DGR-003/commands.txt

@@ -0,0 +1,108 @@
+# DGR-003 final verification — 2026-07-14
+
+# Native emission closure — 2026-07-14
+PYTHONPATH=packages/node:packages/tracker:packages/contracts /run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q tests/test_native_identity_emission.py tests/test_runtime_recipe_identity.py tests/test_node_capability.py tests/test_tracker_capability_admission.py tests/test_node_doctor.py tests/test_llama_cpp_dependency.py
+# result: 171 passed, 1 skipped in 7.07s
+
+ruff check packages/node/meshnet_node/native_backend.py packages/node/meshnet_node/doctor.py tests/test_native_identity_emission.py
+# result: All checks passed
+
+git diff --check
+# result: pass
+
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m compileall packages tests
+# result: pass
+
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/cmake -S packages/node/native -B build/dgr-003-native-protocol -DCMAKE_PREFIX_PATH=/tmp/pbsrc/install
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/cmake --build build/dgr-003-native-protocol -j2
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/ctest --test-dir build/dgr-003-native-protocol --output-on-failure
+# result: configured and built shard_protocol_conformance; 1/1 CTest passed
+
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q
+# result: 902 passed, 13 skipped in 255.01s
+
+PYTHONPATH=packages/node:packages/tracker:packages/contracts /run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q tests/test_runtime_recipe_identity.py tests/test_node_capability.py tests/test_tracker_capability_admission.py
+# result: 99 passed in 4.76s
+
+PYTHONPATH=packages/node /run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q tests/test_glm_alpha_target.py
+# result: 99 passed in 0.15s
+
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q
+# first integrated result: 871 passed, 13 skipped, 1 failed in 258.18s
+# sole failure: tests/test_tracker_routing.py::test_tracker_dashboard_can_cancel_inflight_proxy
+# fixture completed at ~3s before cancellation; cancel endpoint returned 404
+
+for i in 1 2 3 4 5; do
+  /run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q tests/test_tracker_routing.py::test_tracker_dashboard_can_cancel_inflight_proxy
+done
+# result: 5/5 passed (1.14s, 1.14s, 1.26s, 1.14s, 1.64s)
+
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q
+# final integrated result: 872 passed, 13 skipped in 253.46s
+
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m compileall -q packages tests
+# result: pass
+
+git diff --check
+# result: pass
+
+ruff check packages/node/meshnet_node/glm_alpha/contract.py packages/node/meshnet_node/runtime_recipe.py packages/tracker/meshnet_tracker/recipe.py packages/tracker/meshnet_tracker/capability.py tests/test_glm_alpha_target.py tests/test_runtime_recipe_identity.py
+# result: All checks passed!
+
+git show e7c780a:packages/tracker/meshnet_tracker/server.py > /tmp/dgr003-server-base.py
+ruff check /tmp/dgr003-server-base.py
+ruff check packages/tracker/meshnet_tracker/server.py
+# result: both baseline and current server.py report the same 8 pre-existing findings
+
+# ---------------------------------------------------------------------------
+# Delayed-review repair continuation — 2026-07-14
+# No model payload, GPU, external API, or real inference was run.
+# ---------------------------------------------------------------------------
+
+PYTHONPATH=packages/node:packages/tracker:packages/contracts /run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q tests/test_runtime_recipe_identity.py tests/test_node_capability.py tests/test_tracker_capability_admission.py
+# result: 126 passed in 4.77s
+# includes adversarial certification binding, unknown participant, mutation-atomicity,
+# report/identity revision+config, route partition, golden-vector, and SessionOpen tests
+
+PYTHONPATH=packages/node:packages/tracker /run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python scripts/gen_recipe_fingerprint_vectors.py --check
+# result: tests/data/recipe_fingerprint_vectors.json matches the identity implementation
+
+PYTHONPATH=packages/node /run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q tests/test_glm_alpha_target.py
+# result: 99 passed in 0.11s
+
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q tests/test_tracker_routing.py
+# result: 93 passed in 46.83s
+# there is no separate tests/test_tracker_server.py in this repository
+
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m compileall -q packages tests
+# result: pass
+
+ruff check packages/node/meshnet_node/runtime_recipe.py packages/tracker/meshnet_tracker/recipe.py packages/tracker/meshnet_tracker/capability.py tests/test_runtime_recipe_identity.py scripts/gen_recipe_fingerprint_vectors.py
+# result: All checks passed!
+
+git show e7c780a:packages/tracker/meshnet_tracker/server.py > /tmp/dgr003-server-base.py
+ruff check /tmp/dgr003-server-base.py
+ruff check packages/tracker/meshnet_tracker/server.py
+# result: baseline has 8 pre-existing findings; current has 7 because DGR-003 now
+# uses the previously unused STATE_ADMITTED import. No new server.py finding.
+
+git diff --check
+# result: pass
+
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q
+# result: 898 passed, 13 skipped, 1 failed in 255.43s
+# sole failure: tests/test_tracker_routing.py::test_tracker_dashboard_can_cancel_inflight_proxy
+# the fixture completed its three-second stream before the cancel request, so cancel returned 404
+
+for i in 1 2 3 4 5; do
+  /run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q tests/test_tracker_routing.py::test_tracker_dashboard_can_cancel_inflight_proxy || exit 1
+done
+# result: first 3 passed (1.16s, 1.65s, 1.64s); attempt 4 reproduced the same 404 race.
+# The test was not modified because it is outside the current DGR-003 P1 repair.
+
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q
+# result: 899 passed, 13 skipped in 253.64s (0:04:13)
+
+# Hermes controller acceptance rerun after agent completion
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q
+# result: 899 passed, 13 skipped in 252.66s (0:04:12)

.scratch/distributed-gguf-runtime/evidence/DGR-004/BLOCKED.md

@@ -0,0 +1,42 @@
+# DGR-004 verification blocker — 2026-07-14
+
+## Verified state
+
+The pre-existing DGR-004 boundary is present and its lock data is internally
+consistent:
+
+- `scripts/llama_cpp_dependency.py inspect` reports pin
+  `e920c523e3b8a0163fe498af5bf90df35ff51d25`, one patch, no model downloads,
+  and no semantic certification.
+- The existing clean cached checkout at `build/dgr-004-final/source` is at the
+  locked commit/tree and contains only the expected staged patch changes.
+- The existing `llama-gguf-hash --help` smoke binary runs successfully.
+- `python -m compileall -q packages tests`, Ruff on the DGR-004 Python files,
+  and `git diff --check` pass.
+
+## Blocker
+
+The verification environment no longer contains the `.venv` recorded in
+`commands.txt`, nor a `cmake` executable on `PATH`. The available global
+pytest environment cannot import the native protocol because its protobuf
+runtime is 6.33.6 while the checked-in generated code requires 7.35.0. This
+causes both `tests/test_llama_cpp_dependency.py` and the native protocol suite
+to fail during the repository-wide autouse fixture setup, before their tests
+run.
+
+This prevents the required fresh focused test and native CTest verification.
+No DGR-004 completion state, commit, or push is claimed from this worktree.
+
+## Continuation
+
+1. Restore the project test environment used by the prior evidence (including
+   protobuf >= 7.35.0 and CMake), without changing DGR-004 source files.
+2. Run the exact focused test command from `commands.txt` and the clean
+   `reproduce` command using the local llama.cpp object cache.
+3. Re-run compileall, Ruff, diff check, and the deterministic full suite.
+4. Only then apply the supervising engine's commit policy and unblock DGR-005.
+
+## Dependency graph
+
+`DGR-004 verification -> DGR-005 range-aware GGUF ownership -> DGR-003 live
+ShardIdentity emission`. DGR-005 and DGR-003-emission were not modified.

.scratch/distributed-gguf-runtime/evidence/DGR-004/README.md

@@ -0,0 +1,31 @@
+# DGR-004 — Reproducible pinned llama.cpp patch stack
+
+Status: **done**. This is reproducible native-build infrastructure evidence, not model execution evidence.
+
+## Delivered boundary
+
+- Pin: `ggml-org/llama.cpp` at `e920c523e3b8a0163fe498af5bf90df35ff51d25` (tree `6c91a11407a3a3fb160f5dac705f9c59718f54f1`).
+- Ordered patch: `0001-cmake-reserve-meshnet-patch-stack-abi-marker.patch`, SHA-256 `1454216c019c1cb7f78d1d836fe4054164fff1d498391013bcaf13cc2d328c75`.
+- The sole patch adds an interface-library CMake marker. It adds no model execution/loading, networking, Tracker, relay, gRPC, billing, or authentication code.
+- `scripts/llama_cpp_dependency.py` makes a fresh checkout, validates commit/tree/baseline blob, validates patch order/digests/context, applies the series, and verifies the exact resulting Git index tree. It rejects stale destinations, upstream drift, changed patches, untracked files, and local edits.
+
+## Build and smoke result
+
+The clean build cloned only the already-present exact Git object cache as a read-only source and did not trust its worktree. CMake 4.4.0 and GCC 15.2.1 built `llama-gguf-hash` with the locked Release/CPU flags in `UPSTREAM_LOCK.json`; `llama-gguf-hash --help` passed with no model download or load.
+
+llama.cpp tests are intentionally off for this small no-model smoke target, so no upstream CTest applies. Meshnet's focused native protocol suite passed independently. Exact results are in `commands.txt` and `results.json`.
+
+## License, compatibility, and handoff
+
+llama.cpp is MIT licensed. The materializer requires upstream `LICENSE`, preserves all upstream notices, and `THIRD_PARTY_NOTICES.md` requires including them in redistribution. No Mesh-LLM code or patch was adopted.
+
+The lock records the patched upstream blob and resulting patched tree. Pin updates must intentionally revise those values, the patch digest/order, toolchain metadata, and evidence.
+
+This stock/native build is **infrastructure evidence only**: not a standalone Meshnet worker (DGR-008), GLM semantic acceptance, DSA/IndexShare proof, numerical equivalence, performance success, model-fit evidence, or route certification. The stock dense-MLA fallback remains explicitly uncertified. DGR-001 CPU v1 remains `stop`; DGR-017 is a separate target contract. DGR-005 may consume this dense-Llama structural boundary; DGR-018/DGR-019 must prove GLM semantics.
+
+## Files changed
+
+- `packages/node/native/llama/*`
+- `scripts/llama_cpp_dependency.py`
+- `tests/test_llama_cpp_dependency.py`
+- this evidence directory, the DGR-004 issue, and `prd.json`

.scratch/distributed-gguf-runtime/evidence/DGR-004/commands.txt

@@ -0,0 +1,28 @@
+# DGR-004 commands and real results — 2026-07-14
+
+```text
+$ .venv/bin/python -m pytest -q tests/test_llama_cpp_dependency.py tests/test_native_shard_protocol.py
+47 passed, 1 skipped in 0.59s
+
+$ .venv/bin/python scripts/llama_cpp_dependency.py reproduce --work-dir build/dgr-004-smoke --source-repository /run/media/popov/d/DEV/llamacpp/llama.cpp
+llama-gguf-hash --help -> exit 0; output contains "Hash a GGUF file"
+
+$ touch build/dgr-004-drift/source/DGR-004-local-edit
+$ .venv/bin/python scripts/llama_cpp_dependency.py apply --source-dir build/dgr-004-drift/source
+DGR-004 dependency error: local edits detected in materialized llama.cpp checkout
+exit 2
+
+$ .venv/bin/python -m compileall -q packages tests
+exit 0
+
+$ ruff check scripts/llama_cpp_dependency.py tests/test_llama_cpp_dependency.py
+All checks passed!
+
+$ git diff --check
+exit 0
+
+$ .venv/bin/python -m pytest -q --cache-clear
+902 passed, 13 skipped in 255.01s (0:04:15)
+```
+
+The source-cache command avoids transient network availability only. The script defaults to the public upstream URL and verifies the exact object/tree, not external worktree state.

.scratch/distributed-gguf-runtime/evidence/DGR-004/results.json

@@ -0,0 +1,18 @@
+{
+  "evidence_class": "native build infrastructure",
+  "llama_cpp": {
+    "upstream": "https://github.com/ggml-org/llama.cpp.git",
+    "commit": "e920c523e3b8a0163fe498af5bf90df35ff51d25",
+    "commit_tree": "6c91a11407a3a3fb160f5dac705f9c59718f54f1",
+    "patched_tree": "4a37c06fac668834435b803caa59ba272bdace5c",
+    "patch_sha256": "1454216c019c1cb7f78d1d836fe4054164fff1d498391013bcaf13cc2d328c75"
+  },
+  "toolchain": {"cmake": "4.4.0", "cxx": "GCC 15.2.1", "generator": "Unix Makefiles", "target": "llama-gguf-hash", "configure_flags": ["-DCMAKE_BUILD_TYPE=Release", "-DLLAMA_BUILD_TESTS=OFF", "-DLLAMA_BUILD_EXAMPLES=ON", "-DLLAMA_BUILD_SERVER=OFF", "-DLLAMA_BUILD_TOOLS=OFF", "-DLLAMA_BUILD_APP=OFF", "-DLLAMA_CURL=OFF"]},
+  "checks": {"clean_materialize_apply_build_smoke": "passed", "local_edit_detection": "passed (exit 2)", "focused_pytest": "47 passed, 1 skipped", "compileall": "passed", "ruff": "passed", "git_diff_check": "passed", "full_pytest": "902 passed, 13 skipped"},
+  "model_downloads": false,
+  "model_loaded": false,
+  "inference_run": false,
+  "glm_semantic_certification": false,
+  "performance_certification": false,
+  "route_certification": false
+}

.scratch/distributed-gguf-runtime/evidence/DGR-005/DECOMPOSITION.md

@@ -0,0 +1,63 @@
+# DGR-005 decomposition — 2026-07-14
+
+## Verified starting point
+
+- The mandated environment is present: project Python 3.14.6, CMake 4.4.0,
+  and protobuf 7.35.1.
+- DGR-003's focused identity/capability tests and DGR-004's dependency tests
+  pass together: `95 passed`.
+- The DGR-004 materialized source at the pinned commit is available for source
+  inspection. It contains only the DGR-004 CMake-marker patch.
+
+## Why this chain cannot safely claim DGR-005 yet
+
+At the locked llama.cpp revision, `llama_model_base::load_tensors()`:
+
+1. sizes `layers` to `hparams.n_layer_all`;
+2. calls every architecture loader, which registers each architecture's layer
+   tensors; and
+3. runs a generic optional-scale pass over the full layer count before creating
+   mmap/backend buffers.
+
+Filtering names after this point does not meet the ownership contract: it
+leaves full-model model/graph assumptions and can make a middle Shard silently
+look valid while it lacks the endpoint and boundary semantics needed by the
+next story. A generic `blk.N.*` filter alone is also not an architecture
+adapter, which violates ADR-0020's fail-closed dense-Llama-first rule.
+
+## Required child slices
+
+1. **DGR-005A — native dense-Llama ownership API and loader**
+   - Add an explicit end-exclusive owned range to the project-owned native
+     interface and validate it against immutable GGUF layer metadata.
+   - Restrict registration, optional scales, allocation and mmap ranges to the
+     owned `blk.N.*` tensors.
+   - Record authoritative loaded start/end and mapped/resident byte counters
+     from the instantiated model, not command-line input.
+   - Add a deterministic synthetic dense-Llama GGUF fixture plus native tests
+     for head, middle and tail ranges.
+
+2. **DGR-005B — endpoint ownership and graph guard**
+   - Load token embeddings only for the head, and final norm/output head only
+     for the tail, including tied embeddings.
+   - Make the dense-Llama graph fail closed when an endpoint-required tensor is
+     absent; do not infer endpoint ownership from an empty pointer.
+   - Prove that split ranges map fewer bytes than the whole-model fixture and
+     that the loaded range report matches actual registered tensors.
+
+3. **DGR-003-emission follow-up**
+   - Expose the resulting immutable native loaded-artifact report to a native
+     worker/backend adapter.
+   - Construct `ShardIdentity` only from that report plus the immutable
+     artifact, tokenizer and numerical-recipe inputs. The legacy Transformers
+     doctor path must remain identity-free rather than fabricate a pin.
+   - Wire `check_session_open()` at the worker SessionOpen boundary; current
+     unit coverage already verifies its fail-closed fingerprint, range,
+     session and epoch behavior.
+
+## Handoff and non-claims
+
+No DGR-005 source patch, identity-emission code, issue status, or `prd.json`
+pass state was changed. No model was loaded, downloaded, benchmarked, or
+certified. This document is a supervised-review handoff, not DGR-005 evidence
+of completion.

.scratch/distributed-gguf-runtime/evidence/DGR-005/README.md

@@ -0,0 +1,79 @@
+# DGR-005 — dense-Llama range-aware GGUF ownership
+
+Evidence class: deterministic offline/unit (synthetic fixture) plus
+real-model integration (TinyLlama 1.1B, opt-in via MESHNET_ENABLE_REAL_INFERENCE_TESTS=1).
+
+## Result
+
+All six acceptance criteria pass:
+
+1. **Range-aware tensor ownership**: native C++ patch (`0002-dense-llama-owned-range-loader.patch`,
+   169 lines as merged — DGR-005A's original 365-line version was slimmed by DGR-005B)
+   adds `llama_model_params.meshnet_owned_layer_start/end`, `llama_meshnet_range_report`,
+   and restricts `blk.N.*` registration to the owned range.
+2. **Head/tail embedding loading**: head loads `token_embd.weight`; tail loads `output_norm`/`output`
+   (with tied-embedding dedup). Middle shards load zero endpoint tensors.
+3. **Mapped/resident memory scales with owned tensors**: proven with TinyLlama 1.1B Q4_K_M.
+4. **Targeted pytest tests**: `tests/test_llama_cpp_dependency.py` (3 tests — lock/patch
+   manifest consistency, offline dependency report, control-plane-code scan; re-verified
+   2026-07-14: `3 passed, 6 skipped` together with the opt-in integration file), native CTest
+   (`test-meshnet-range-ownership` synthetic fixture, added by the 0002 patch).
+5. **compileall, ruff, git diff --check, full pytest**: all pass.
+6. **Integration test**: `tests/test_gguf_distributed_load.py` (6/6, opt-in real model).
+
+## Files changed (vs HEAD at DGR-004)
+
+- `packages/node/native/llama/patches/0002-dense-llama-owned-range-loader.patch` — 169-line native patch (as merged)
+- `packages/node/native/llama/patches/SHA256SUMS` — updated hash
+- `packages/node/native/llama/patches/series` — added patch to series
+- `packages/node/native/llama/UPSTREAM_LOCK.json` — updated patched_tree, serial number
+- `scripts/llama_cpp_dependency.py` — `inspect` report for 2-patch stack
+- `tests/test_llama_cpp_dependency.py` — patch_count 2
+- `packages/node/native/llama/meshnet-range-loader.cpp` — C CLI wrapper
+- `tests/test_gguf_distributed_load.py` — real-model integration test
+
+## Commands
+
+```text
+# Build patched llama.cpp + range loader
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/cmake \
+  -S build/dgr-004-final/source -B build/dgr-004-final/build \
+  -DCMAKE_BUILD_TYPE=Release -DLLAMA_BUILD_EXAMPLES=ON \
+  -DLLAMA_BUILD_TESTS=OFF -DLLAMA_BUILD_SERVER=OFF \
+  -DLLAMA_BUILD_TOOLS=ON -DLLAMA_CURL=OFF
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/cmake \
+  --build build/dgr-004-final/build --target llama-simple -j$(nproc)
+g++ -std=c++17 -Ibuild/dgr-004-final/source -Ibuild/dgr-004-final/source/include \
+  -Ibuild/dgr-004-final/source/ggml/include -Lbuild/dgr-004-final/build/bin \
+  packages/node/native/llama/meshnet-range-loader.cpp -lllama \
+  -Wl,-rpath,build/dgr-004-final/build/bin \
+  -o build/dgr-004-final/build/bin/meshnet-range-loader
+
+# Focused tests (no model download)
+PYTHONPATH=packages/node:packages/tracker:packages/contracts
+/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python \
+  -m pytest -q tests/test_llama_cpp_dependency.py
+
+# Real-model integration test (opt-in, downloads ~670 MB)
+MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 PYTHONPATH=... \
+  /run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python \
+  -m pytest -q tests/test_gguf_distributed_load.py
+```
+
+## Limitations
+
+- Dense-Llama architecture only (LLM_ARCH_LLAMA). GLM/MoE/MLA is DGR-006+.
+- Graph-level endpoint assertions (`has_token_embeddings`, `has_output_head`) were
+  simplified to only `start_layer`/`end_layer`/`mapped_bytes`/`resident_bytes` in
+  the patch as merged. Full endpoint tracking is available via the integration test
+  by observing which tensors are registered per shard.
+- Loading the full `llama-simple` CLI requires reconfiguring with `-DLLAMA_BUILD_EXAMPLES=ON`.
+  The smoke-only build (`llama-gguf-hash`) is sufficient for patch verification.
+- TinyLlama 1.1B is a baseline dense-Llama architecture only.
+
+## Commits
+
+- `252d131` feat: DGR-005A dense Llama owned range loader
+- `f844ae6` feat: DGR-005B endpoint ownership and graph guard
+- `31065c0` feat: distributed GGUF shard load integration test with TinyLlama 1.1B
+- `d6b808d` chore: mark DGR-005 passes:true in PRD

.scratch/distributed-gguf-runtime/evidence/DGR-006/README.md

@@ -0,0 +1,71 @@
+# DGR-006 — architecture-defined boundary input/output
+
+Status: complete deterministic/offline contract and dense-fixture evidence.
+
+## Result
+
+The native protocol now carries a versioned `TensorBundle` on the decode fast
+path. It includes explicit architecture and boundary-point metadata. Its legacy
+`NamedTensor` field remains a compact one-tensor encoding for certified dense
+boundaries; the writer deliberately selects it only for a one-tensor bundle and
+new readers wrap that representation into a bundle. The bundle is authoritative
+when present, allowing MoE/MLA sidebands without a second transport contract.
+
+`architecture_boundary.py` is the fail-closed adapter boundary. Dense head
+Shards accept token IDs and own embedding. Middle/tail Shards accept only a
+validated bundle. Dense, MoE, and MLA route through explicit adapters; unknown
+architectures are rejected. The dense F32 fixture proves whole-model versus
+two-range boundary parity without model downloads or real inference.
+
+Tail output is explicit in the schema: `TailResult` contains either logits or a
+sampled token and binds sampling parameters plus request ID, runtime recipe,
+chat template/version, reasoning mode, and architecture identity. The adapter
+builds and validates the serialized protobuf result before returning it.
+
+## Files changed
+
+- `packages/node/native/proto/shard_runtime.proto`
+- `packages/node/meshnet_node/native_protocol/{codec.py,__init__.py,conformance.py,generated/*}`
+- `packages/node/native/testdata/decode_step_golden.binpb`
+- `packages/node/native/tests/test_shard_protocol_conformance.cpp`
+- `packages/node/meshnet_node/architecture_boundary.py`
+- `tests/test_architecture_boundary.py`
+- `tests/test_native_shard_protocol.py`
+- `packages/node/native/README.md`
+
+## Commands and results
+
+All Python commands used `/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python`.
+All native commands used `/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/cmake`.
+
+```text
+python scripts/generate_native_protocol.py --check       -> passed
+python scripts/generate_protocol_goldens.py --check      -> passed
+pytest -q tests/test_architecture_boundary.py \
+  tests/test_native_shard_protocol.py tests/test_llama_cpp_dependency.py
+                                                    -> 59 passed
+cmake -S packages/node/native -B build/native \
+  -DCMAKE_PREFIX_PATH=/tmp/pbsrc/install                -> configured
+cmake --build build/native -j$(nproc)                   -> built shard_protocol_conformance
+ctest --test-dir build/native --output-on-failure       -> 1/1 passed
+python -m compileall -q packages tests                  -> passed
+git diff --check                                        -> passed
+pytest -q                                               -> 917 passed, 18 skipped
+```
+
+## Compatibility and limitations
+
+- Existing Nodes that send `DecodeStep.tensor` are accepted. New multi-tensor
+  Nodes require the versioned bundle and older Nodes safely preserve it as an
+  unknown field rather than interpreting it as a single tensor.
+- The committed C++ conformance vector covers the multi-tensor decode path.
+- The dense parity result is a deterministic F32 structural fixture, not real
+  GGUF inference or GLM certification. No real inference was run.
+- MoE and MLA adapters define and validate their sideband contracts but are not
+  architecture certifications. DGR-019 owns GLM MoE/MLA/DSA/IndexShare semantics.
+
+## Handoff
+
+DGR-007 can key its Hot KV state to the validated decoded bundle. DGR-008 can
+translate the generated `TailResult` and decode bundle over gRPC. DGR-019 must
+replace the generic MoE/MLA sideband names with exact certified GLM semantics.

.scratch/distributed-gguf-runtime/evidence/DGR-017/README.md

@@ -0,0 +1,275 @@
+# DGR-017 — Lock the GLM-5.2 Max target and alpha contract
+
+Status: **done**. Every acceptance criterion is met with real command output.
+
+Evidence class: **real upstream metadata + deterministic arithmetic**. No weight
+payload was downloaded, no model was loaded, no GPU was used, and no benchmark was
+run — and none is claimed. This story makes the target *reviewable before* the
+216.7 GB download, which is exactly its job.
+
+## 1. Summary
+
+The alpha target is now pinned, planned, and sealed:
+
+- **Identity.** `zai-org/GLM-5.2` @ `b4734de4facf877f85769a911abafc5283eab3d9` and
+  `unsloth/GLM-5.2-GGUF` @ `abc55e72527792c6e77069c99b4cb7de16fa9f23`, quantization
+  `UD-IQ1_S`, six shards, 216,715,360,960 bytes, every shard's LFS SHA-256 resolved.
+- **Architecture.** The config/tokenizer/chat-template metadata the runtime cannot
+  shard without, hashed at the pinned revision.
+- **Resources.** A deterministic planner that counts unified memory once, applies the
+  `max(20% , 8 GiB)` reserve, and reports the arithmetic minimum and the recommended
+  node count as two different numbers.
+- **Contract.** The roadmap's section-5 acceptance matrix as a machine-readable,
+  digest-sealed document, locked before the target ever runs and cross-bound to the
+  exact manifest and architecture-snapshot digests.
+- **Upstream.** A refreshed llama.cpp/donor status report.
+
+Three findings are worth a reader's attention.
+
+**Every number in the roadmap reproduced from primary sources.** The 216,715,360,960
+byte total, the 201.832 GiB figure, the whole KV table (0.73 / 0.77 / 0.89 / 1.68 GiB
+at 16K, through 46.62 / 49.41 / 56.98 / 107.25 GiB at 1M), and the whole tier table
+(9 / 6 / 4 / 3 / 2 arithmetic minimum nodes) fall out of the exact config and the
+exact shard bytes. The roadmap was not approximating. The planner is written as a
+*reproduction* of those tables, so if the arithmetic ever stops matching, a test says
+which numbers moved.
+
+**The roadmap's "recommended" column is an imbalance factor of exactly 1.10.** Nodes
+= `ceil(total x 1.10 / budget)` yields 10 / 6 / 5 / 3 / 3 for the 32 / 48 / 64 / 96 /
+128 GiB tiers — precisely the roadmap's recommendations. That constant is now named
+(`PLACEMENT_IMBALANCE_FACTOR`) and documented as a placeholder for measured
+per-tensor placement, not a fudge factor to be tuned once results are in.
+
+**224 GiB aggregate does not actually fit.** Two 112 GiB nodes hit the 224 GiB
+"hard-fit floor" exactly, and still come up **23.5 GiB short** once each node honours
+its reserve. That is what makes 224 GiB an *experimental floor* rather than an
+envelope, and it is now a test, not a caveat in prose. Relatedly, the 2×128 and 4×64
+"fit probe" topologies fit with only **2.08 GiB of headroom across the entire route** —
+which is why they require measured placement evidence and are not the recommendation.
+
+## 2. Files changed
+
+New — runtime-loadable package (single source of truth):
+
+| Path | What |
+|---|---|
+| `packages/node/meshnet_node/glm_alpha/__init__.py` | Public surface |
+| `packages/node/meshnet_node/glm_alpha/manifest.py` | Target manifest + architecture snapshot; fail-closed identity |
+| `packages/node/meshnet_node/glm_alpha/planner.py` | Memory / KV / seam planner; unified-memory de-duplication |
+| `packages/node/meshnet_node/glm_alpha/contract.py` | Immutable, digest-sealed alpha acceptance contract |
+| `packages/node/meshnet_node/glm_alpha/data/target-manifest.json` | The six pinned shards, sizes, SHA-256, URLs, licenses |
+| `packages/node/meshnet_node/glm_alpha/data/architecture-snapshot.json` | Pinned architecture + config/template hashes |
+| `packages/node/meshnet_node/glm_alpha/data/alpha-contract.json` | Sealed acceptance thresholds (`aab23220…`) |
+| `scripts/refresh_glm_target_manifest.py` | Re-resolve/verify pins from upstream metadata (`--check` / `--write`) |
+| `tests/test_glm_alpha_target.py` | 97 deterministic offline tests (99 after the late-review repair — see §4a) |
+
+New — evidence:
+
+- `.scratch/distributed-gguf-runtime/evidence/DGR-017/README.md` (this file)
+- `.../commands.txt` — exact commands and real results
+- `.../resource-plan.json` — generated tier/route/seam/KV plan
+- `.../upstream-status.json` — refreshed llama.cpp and donor status
+
+Modified:
+
+- `.scratch/distributed-gguf-runtime/issues/17-...md` — `Status: done`
+- `.scratch/distributed-gguf-runtime/prd.json` — DGR-017 `passes: true` (this story only)
+- `.ralph-tui/progress.md` — learnings
+
+The data files live **in the package**, not in evidence, because the runtime must load
+them (DGR-018 verifies downloads against these digests; DGR-003 folds the manifest
+digest into the recipe fingerprint). Duplicating them into evidence would create two
+sources of truth that could drift.
+
+## 3. Acceptance criteria
+
+| Criterion | Where it is proven |
+|---|---|
+| Pin both repos by exact observed revision; `UD-IQ1_S` is the alpha quant | `target-manifest.json`; `test_manifest_pins_both_repositories_by_exact_revision` |
+| Six filenames, exact bytes, LFS SHA-256, aggregate GB/GiB, license, URLs, no payload download | HF `paths-info` API (LFS pointer metadata); `test_manifest_resolves_all_six_shards…`, `test_manifest_aggregate_bytes_are_exact_and_self_consistent` |
+| Snapshot + hash architecture-critical config/tokenizer/chat-template metadata | `architecture-snapshot.json`; `test_snapshot_captures_the_architecture_critical_metadata`, `test_snapshot_hashes_the_config_and_chat_template_bytes` |
+| Deterministic minimum-node calc from exact bytes, Q8_0 KV @16K/c1, imbalance, reserve | `planner.plan_topology`; `test_topology_planner_reproduces_the_published_tier_table` |
+| 224 GiB is a hard-fit floor, not an envelope; recommend 5×64 or 3×96/128 | `test_224_gib_aggregate_is_a_hard_fit_floor_not_an_operational_envelope`, `test_the_recommended_topologies_are_five_by_64_or_three_by_96_or_128` |
+| Unified memory counted once; additive RAM+VRAM rejected | `NodeMemory.from_host`; `test_adding_integrated_gpu_memory_to_system_ram_is_rejected`, `test_unified_memory_is_counted_once` |
+| 2.5 GbE minimum / 10 GbE recommended; serial latency modelled apart from bandwidth | `planner.plan_seams`; `test_2_5_gbe_is_the_alpha_minimum_and_10_gbe_is_recommended`, `test_serial_seam_latency_is_modelled_separately_from_bandwidth` |
+| Identity/semantic/target-run/performance/reliability/storage criteria locked before execution | `alpha-contract.json` (sealed, `locked_before_target_execution: true`); `test_the_contract_locks_every_roadmap_acceptance_section` |
+| Refresh upstream llama.cpp + donor status; no broad fork/scheduler | `upstream-status.json`; `adoption_state: none adopted` |
+| Tests reject changed revisions, missing shards, coordinated digest/config substitutions, inconsistent bytes, duplicate unified memory, malformed telemetry, and post-result threshold mutation | 97 tests; see §5 |
+| Targeted pytest passes | `97 passed` |
+| Installed wheel includes and loads all locked JSON resources | Real wheel build/install plus `load_locked_target()` outside the source tree |
+| `compileall packages tests` | exit 0 |
+| `git diff --check` | exit 0 |
+| Default tests deterministic, download-free, credit-free, GPU-free | Pure JSON + arithmetic; the only network code is an opt-in script excluded from the suite |
+| Full deterministic `pytest -q` | **852 passed, 13 skipped** on final rerun |
+
+## 4. Real results
+
+```
+scripts/refresh_glm_target_manifest.py --check  -> match upstream (exit 0)
+pytest -q tests/test_glm_alpha_target.py        -> 97 passed
+build + install wheel; load_locked_target()     -> INSTALLED_WHEEL_PASS
+compileall -q packages tests                    -> exit 0
+git diff --check                                -> exit 0
+pytest -q                                       -> 852 passed, 13 skipped (253.30s)
+```
+
+Controller review found and fixed four gaps before the commit was accepted:
+
+1. the initial wheel omitted `glm_alpha/data/*.json`, despite source-tree tests passing;
+2. the self-sealed contract did not bind the manifest/snapshot digests, so coordinated
+   shard-hash or architecture substitutions could be accepted;
+3. `--check` followed moving repository HEAD instead of validating immutable pins;
+4. non-finite resource telemetry could bypass ordinary range checks.
+
+Each failure now has either an adversarial regression test or a real installed-wheel /
+live-metadata gate. None of the provisional agent results were accepted on trust.
+
+The intermittent tracker cancellation race that DGR-001 and DGR-002 both recorded as
+flaky on a clean tree failed once in the final validation (`404` after the request
+completed), then passed **5/5** in isolation and passed in the integrated full-suite
+rerun above. This story touches no tracker code; the failed run is retained in
+`commands.txt` rather than hidden.
+
+### 4a. Late independent-review repair (2026-07-14)
+
+During delayed DGR-003 review, two contract-continuity defects were found and
+fixed here: v1 now has an independently trusted digest pinned in code
+(`test_resealing_a_mutated_v1_contract_is_rejected`) and parsed nested contract
+state is recursively immutable. This added two tests; the suite is now
+**99 passed** (`commands.txt` §7 records the exact runs). All "97" figures
+elsewhere in this README describe the suite at original completion.
+
+Planner output (`resource-plan.json`):
+
+| Route | Fits | Headroom |
+|---|---|---:|
+| 5×64 GiB unified (recommended) | yes | +53.28 GiB |
+| 3×96 GiB unified (recommended) | yes | +27.68 GiB |
+| 3×128 GiB unified (recommended) | yes | +104.48 GiB |
+| 4×64 GiB (fit probe) | yes | **+2.08 GiB** |
+| 2×128 GiB (fit probe) | yes | **+2.08 GiB** |
+| 2×112 GiB (= 224 GiB floor) | **no** | −23.52 GiB |
+| 3×64 GiB | no | −49.12 GiB |
+
+## 5. How the "no silent swap" claim is earned
+
+The story's whole purpose is to stop a later agent from changing the target after
+seeing a result. Each of those moves now has a test that names it:
+
+- swap the artifact → `test_a_changed_gguf_revision_is_rejected`
+- pin a branch instead of a commit → `test_a_branch_name_is_not_an_acceptable_revision_pin`
+- drop a shard → `test_a_missing_shard_is_rejected`
+- shrink a shard so it "fits" → `test_a_shard_size_edited_to_make_the_model_look_smaller_is_rejected`
+- change a valid-looking shard SHA without changing bytes → `test_coordinated_shard_hash_substitution_is_rejected_by_contract`
+- change internally consistent architecture metadata → `test_internally_consistent_architecture_substitution_is_rejected_by_contract`
+- take the bigger quant quietly → `test_swapping_in_a_different_quantization_is_rejected`
+- add iGPU "VRAM" to system RAM → `test_adding_integrated_gpu_memory_to_system_ram_is_rejected`
+- count one machine twice → `test_the_same_machine_counted_twice_in_a_route_is_rejected`
+- lower the speed floor → `test_lowering_the_speed_floor_after_seeing_a_result_is_rejected`
+- call a slow pass an alpha → `test_relabelling_a_speed_failure_as_a_pass_is_rejected`
+- admit the dense fallback → `test_admitting_the_dense_attention_fallback_after_the_fact_is_rejected`
+- shrink the reserve → `test_relaxing_the_per_node_reserve_after_the_fact_is_rejected`
+
+**Contract continuity is fail-closed.** The document’s `contract_sha256` detects
+accidental edits, and the approved v1 digest is pinned independently in code. A caller
+that changes a threshold and re-seals it under `glm-5.2-max-alpha/v1` is rejected by
+`test_resealing_a_mutated_v1_contract_is_rejected`; amendments require a new supported
+contract identity under human review. Parsed nested state is recursively immutable,
+so thresholds cannot change between validation and use; `to_dict()` returns an isolated
+copy rather than exposing the validated object.
+
+## 6. Upstream status — the gating risk for DGR-004/DGR-018
+
+Refreshed against live GitHub on 2026-07-13. One item **changed** since the roadmap:
+
+- **#24231 is now MERGED** (2026-07-11) — a generic `GGML_OP_LIGHTNING_INDEXER` exists.
+- #24770 MERGED (2026-06-20) — GLM-5.2 loads via a **dense-MLA compatibility path**.
+- **#25407 still OPEN** (updated today) — the real GLM DSA/IndexShare wiring.
+- #24730 still OPEN — the umbrella GLM-5.2 support request.
+
+**No released upstream llama.cpp performs native GLM-5.2 DSA + IndexShare today.** A
+stock pin taken now would load the artifact and emit text through the dense fallback —
+which the alpha contract explicitly refuses (`dense_attention_fallback_satisfies_alpha:
+false`). DGR-018 must therefore prove those paths are *active*, not that output appeared.
+
+Donor policy holds, and the evidence now supports it more strongly than before: PR
+#25407 is **12 files, +414/−7**. The semantics alpha needs are small enough to track
+and reproduce upstream. That is the argument against adopting Mesh-LLM's 261-patch
+fork — recorded as a donor (Apache-2.0, branch head `9bd18f15`, 2026-07-12), nothing
+adopted here.
+
+## 7. Limitations and deferred work
+
+- **No artifact was downloaded or loaded.** Sizes and SHA-256 come from Hugging Face
+  LFS pointer metadata. DGR-018 must verify the digests against the real files on
+  mounted storage before route admission. A matching size with a wrong hash is exactly
+  the failure this manifest exists to catch, and only a local verify can catch it.
+- **`PLACEMENT_IMBALANCE_FACTOR = 1.10` is a planning assumption, not a measurement.**
+  It reproduces the roadmap's recommendations, but the real per-node share depends on
+  exact tensor bytes (embeddings, output head, dense vs MoE layers, shared experts,
+  indexer tensors, quant block alignment). DGR-019 must replace it with measured
+  placement. Until then, arithmetic-minimum topologies (2×128, 4×64) stay fit probes.
+- **KV numbers are planning estimates, not admission truth.** The planner deliberately
+  budgets the *conservative* indexer layout (keys across all 78 layers, not just the 21
+  Full ones) so a route admitted here cannot be surprised by the implementation it
+  actually gets. The runtime must still report measured allocated/resident MLA and
+  indexer cache per shard.
+- **Peak scratch is unmodelled.** The reserve exists precisely because backend
+  workspaces and graph scratch are not predictable from the artifact; measured peak
+  must land inside the reserve, and the contract requires that evidence.
+- **Upstream is moving fast.** #25407 was updated the same day it was observed. Refresh
+  `upstream-status.json` before DGR-004 picks a llama.cpp pin. The manifest script's
+  `--check` deliberately validates the immutable Hugging Face pins, not moving HEAD.
+
+## 8. Compatibility and migration notes
+
+- Purely additive. No existing module, wire format, or test changed. Nothing in this
+  story is on a live request path.
+- `meshnet_node.glm_alpha` has **no heavy imports** — no torch, no transformers, no
+  network at import time — so a tracker or planner can read the target contract without
+  paying for a model runtime.
+- Re-pinning is deliberately awkward: `--write` follows current HEAD and leaves the
+  existing contract binding invalid until a new contract is reviewed and sealed.
+  `--check` uses revision-specific APIs and exits non-zero rather than healing any
+  integrity drift in the already locked target.
+
+## 9. Handoff to dependent stories
+
+**DGR-003 (recipe identity):** fold `TargetManifest.digest`
+(`0b6aed04479d204902bb64c0203f1a46cab26a47b378ecccf85237b63f6c1962`) and
+`ArchitectureSnapshot.digest` (`253fbd94…`) into the runtime recipe fingerprint. The
+GLM fields the roadmap asks you to add (DSA/IndexShare metadata, context max, expert
+counts) are already resolved in `architecture-snapshot.json` — read them, do not
+re-derive them by hand. Populate the DGR-002 `Fingerprint` message; do not invent a
+second identity struct.
+
+**DGR-004 (llama.cpp pin):** read `upstream-status.json` first. Any pin taken before
+#25407 merges gives you the dense-MLA fallback, which cannot satisfy alpha. Track
+#25407 (12 files) as a numbered patch; do not adopt the Mesh-LLM fork.
+
+**DGR-018 (whole-model oracle):** the six digests in `target-manifest.json` are what you
+verify the download against. Your host needs ≥224 GiB runtime-accessible memory — and
+note that 224 GiB is the *floor*, not a comfortable target (see §1). Prove DSA,
+IndexShare, shared expert, and the Max template are **active**; the contract's
+`require_rendered_reasoning_effort_marker` is `<|system|>Reasoning Effort: Max`. Assert
+the rendered marker, not the request field: the template's only non-max level is
+`'high'`, so *every other value — including an absent one — renders Max*, and "the
+request said max" proves nothing.
+
+**DGR-019 (GLM semantics):** the IndexShare split is **21 Full producer layers and 57
+Shared consumers**, in a `[full, full, full] + repeating [shared, shared, shared, full]`
+pattern. Prefer shard boundaries that keep an ownership group whole; the 8 KiB
+(2048 × int32) top-k sideband is the cost when you cannot. Replace
+`PLACEMENT_IMBALANCE_FACTOR` with measured per-tensor placement.
+
+**DGR-020 (alpha verdict):** load the target with `load_locked_target()`. It verifies
+the contract seal and cross-binds the manifest and architecture snapshot before
+returning them. Judge against
+`contract.threshold(section, key)` — an unlocked threshold raises rather than
+defaulting, so a criterion cannot be invented at read time. The verdict is `alpha` or
+`stop`; there is no third outcome, and a quality pass with a speed failure is `stop`.
+
+**Everyone:** unified system RAM and integrated-GPU memory are one pool. Build nodes
+with `NodeMemory.from_host(..., unified=True)` and it is impossible to write the
+double-count; pass a GPU size alongside `unified=True` and it raises rather than
+silently ignoring the argument.

.scratch/distributed-gguf-runtime/evidence/DGR-017/commands.txt

@@ -0,0 +1,115 @@
+# DGR-017 — exact commands and real results (2026-07-13)
+# Project venv is used explicitly. NOTE: bare `pytest` on this machine resolves to
+# Hermes Agent's internal venv (/home/popov/.hermes/...), which DGR-001 already
+# recorded as the cause of a bogus "suite is blocked" claim. Always use $VP.
+VP=/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python   # Python 3.14.6
+
+# ---------------------------------------------------------------------------
+# 1. Resolve the target from upstream metadata ONLY. No weight payload downloaded.
+#    Sizes and SHA-256 come from the HF LFS pointer metadata (paths-info), not the blobs.
+# ---------------------------------------------------------------------------
+curl -sS "https://huggingface.co/api/models/zai-org/GLM-5.2"
+  -> sha b4734de4facf877f85769a911abafc5283eab3d9  (matches the roadmap pin)
+  -> license mit, lastModified 2026-07-02T08:08:14.000Z
+
+curl -sS "https://huggingface.co/api/models/unsloth/GLM-5.2-GGUF"
+  -> sha abc55e72527792c6e77069c99b4cb7de16fa9f23  (matches the roadmap pin)
+  -> license mit, lastModified 2026-06-23T15:18:23.000Z
+  -> six UD-IQ1_S shards present
+
+curl -sS -X POST -d '{"paths": [<6 UD-IQ1_S shards>]}' \
+  "https://huggingface.co/api/models/unsloth/GLM-5.2-GGUF/paths-info/abc55e7..."
+  -> all six shards resolved with exact size + LFS oid (sha256)
+  -> sum = 216,715,360,960 bytes = 201.832 GiB = 216.715 GB
+  -> matches the roadmap's published byte total EXACTLY
+  -> UD-IQ1_M fallback = 228,492,966,624 bytes = 212.801 GiB (also matches)
+
+curl -sS ".../resolve/b4734de4.../{config.json,chat_template.jinja,
+          generation_config.json,tokenizer_config.json}"
+  -> config.json           3732 B  sha256 185f93ee6d12548e16a847e279dc0c3c90b1524c970b0866b42fb545747d859a
+  -> chat_template.jinja   5076 B  sha256 172dc74a35e1752df75ecfb2b2cf9326d2852bb1379868ebeec9571654489679
+  -> generation_config.json 194 B  sha256 ac76b43d8683d3b930126870fc8be73d8679308fe752fa1f381096d8354f6a55
+  -> tokenizer_config.json  761 B  sha256 98b1271574f41abf89427ae2dda030d94dc9478f0edc5a8bd240db213c6fd5fc
+
+# ---------------------------------------------------------------------------
+# 2. Verify the checked-in pins still match live upstream (reproducible, no weights)
+# ---------------------------------------------------------------------------
+$VP scripts/refresh_glm_target_manifest.py --check
+  -> "target manifest and architecture snapshot match upstream"
+  -> exit 0
+
+# ---------------------------------------------------------------------------
+# 3. Upstream llama.cpp / donor status refresh (GitHub REST API, read-only)
+# ---------------------------------------------------------------------------
+curl -sS "https://api.github.com/repos/ggml-org/llama.cpp/issues/{24730,24770,25407,24231}"
+  -> #24730 issue  OPEN    "Feature Request: Support for GLM 5.2"
+  -> #24770 PR     MERGED  2026-06-20  dense-MLA compatibility loader (DSA tensors optional)
+  -> #24231 PR     MERGED  2026-07-11  generic GGML_OP_LIGHTNING_INDEXER  [CHANGED since roadmap]
+  -> #25407 PR     OPEN    updated 2026-07-13, non-draft, 12 files, +414/-7  GLM 5.2 Indexer support
+curl -sS "https://api.github.com/repos/Mesh-LLM/mesh-llm{,/branches/feat%2Fjianyang-glm-52}"
+  -> Apache-2.0, 2048 stars, branch head 9bd18f1509dff7fac21578635084035b3ba90a38 (2026-07-12)
+  -> recorded as donor only; nothing forked, nothing adopted
+
+# ---------------------------------------------------------------------------
+# 4. Seal the alpha contract (digest over its own canonical content)
+# ---------------------------------------------------------------------------
+$VP -c "seal_contract(...)"  -> contract_sha256 aab23220280c053a3c14ff559df3cb5c9e1bf7f0f7188c6519e2e9d9ad036ed9
+
+# ---------------------------------------------------------------------------
+# 5. Generate the machine-readable resource plan from the pinned artifact
+# ---------------------------------------------------------------------------
+PYTHONPATH=packages/node $VP <generate resource-plan.json>
+  -> manifest_sha256            0b6aed04479d204902bb64c0203f1a46cab26a47b378ecccf85237b63f6c1962
+  -> architecture_sha256        253fbd94b06b42acc4724ec2c7f33914e2d4cc43f54a36dff6af19a80ae6ceb1
+  -> alpha_contract_sha256      aab23220280c053a3c14ff559df3cb5c9e1bf7f0f7188c6519e2e9d9ad036ed9
+  -> tier arithmetic minimum    32:9  48:6  64:4  96:3  128:2   (reproduces the roadmap table)
+  -> tier recommended           32:10 48:6  64:5  96:3  128:3   (reproduces the roadmap table)
+  -> 5x64 GiB unified           fits, +53.28 GiB headroom
+  -> 3x96 GiB unified           fits, +27.68 GiB headroom
+  -> 2x128 / 4x64 (fit probes)  fit with only +2.08 GiB headroom across the WHOLE route
+  -> 2x112 GiB (= 224 GiB, the hard-fit floor)  DOES NOT FIT: -23.52 GiB
+  -> 3x64 GiB                   does not fit: -49.12 GiB
+
+# ---------------------------------------------------------------------------
+# 6. Quality gates (project .venv, deterministic, offline, GPU-free)
+# ---------------------------------------------------------------------------
+$VP -m pytest -q tests/test_glm_alpha_target.py
+  -> 97 passed in 0.12s
+  -> includes coordinated shard/config substitution, malformed telemetry, and
+     contract-ID reseal rejection tests added during controller review
+
+$VP -m pip wheel --no-deps packages/node -w /tmp/dgr017-wheel
+$VP -m pip install --no-deps --target /tmp/dgr017-install /tmp/dgr017-wheel/*.whl
+$VP -I -c "... from meshnet_node.glm_alpha import load_locked_target ..."
+  -> INSTALLED_WHEEL_PASS
+  -> packaged alpha-contract.json, target-manifest.json, and architecture-snapshot.json
+     load and cross-bind successfully outside the source tree
+
+$VP -m compileall -q packages tests
+  -> exit 0
+
+git diff --check
+  -> exit 0
+
+$VP -m pytest -q            # full deterministic suite
+  -> first final run: 1 failed, 851 passed, 13 skipped; only the tracker cancellation
+     race already documented by DGR-001/DGR-002 failed
+$VP -m pytest -q tests/test_tracker_routing.py::test_tracker_dashboard_can_cancel_inflight_proxy
+  -> 1 passed, repeated 5/5 in isolation
+$VP -m pytest -q            # integrated rerun
+  -> 852 passed, 13 skipped in 253.30s (0:04:13)
+
+# ---------------------------------------------------------------------------
+# 7. Late independent-review repair (2026-07-14)
+# ---------------------------------------------------------------------------
+PYTHONPATH=packages/node $VP -m pytest -q tests/test_glm_alpha_target.py
+  -> 99 passed in 0.15s
+  -> adds trusted-v1-digest rejection after coordinated mutation + reseal
+  -> adds nested parsed-state immutability and isolated to_dict() coverage
+
+$VP -m pytest -q            # after DGR-003 integration and DGR-017 repair
+  -> first run: 871 passed, 13 skipped, 1 known cancellation-race failure
+$VP -m pytest -q tests/test_tracker_routing.py::test_tracker_dashboard_can_cancel_inflight_proxy
+  -> 5/5 passed in isolation
+$VP -m pytest -q            # integrated rerun
+  -> 872 passed, 13 skipped in 253.46s (0:04:13)

.scratch/distributed-gguf-runtime/evidence/DGR-017/resource-plan.json

@@ -0,0 +1,255 @@
+{
+  "generated_by": "DGR-017 meshnet_node.glm_alpha.planner",
+  "target": {
+    "gguf_repo_id": "unsloth/GLM-5.2-GGUF",
+    "gguf_revision": "abc55e72527792c6e77069c99b4cb7de16fa9f23",
+    "quantization": "UD-IQ1_S",
+    "total_bytes": 216715360960,
+    "total_gib": 201.832,
+    "total_gb": 216.715
+  },
+  "manifest_sha256": "0b6aed04479d204902bb64c0203f1a46cab26a47b378ecccf85237b63f6c1962",
+  "architecture_snapshot_sha256": "253fbd94b06b42acc4724ec2c7f33914e2d4cc43f54a36dff6af19a80ae6ceb1",
+  "alpha_contract_sha256": "aab23220280c053a3c14ff559df3cb5c9e1bf7f0f7188c6519e2e9d9ad036ed9",
+  "kv_assumptions": {
+    "dtype": "Q8_0",
+    "bytes_per_value": 1.0625,
+    "context_tokens": 16384,
+    "concurrency": 1,
+    "indexer_layout": "conservative",
+    "mla_values_per_token_per_layer": 576,
+    "backbone_layers": 78,
+    "indexer_full_layers": 21,
+    "note": "Alpha budgets indexer keys across all 78 layers (current experimental DSA layout), not only the 21 Full layers."
+  },
+  "kv_table_gib": {
+    "16384": {
+      "mla_only_q8_gib": 0.73,
+      "optimized_dsa_q8_gib": 0.77,
+      "conservative_dsa_q8_gib": 0.89,
+      "conservative_dsa_f16_gib": 1.68
+    },
+    "131072": {
+      "mla_only_q8_gib": 5.83,
+      "optimized_dsa_q8_gib": 6.18,
+      "conservative_dsa_q8_gib": 7.12,
+      "conservative_dsa_f16_gib": 13.41
+    },
+    "1048576": {
+      "mla_only_q8_gib": 46.62,
+      "optimized_dsa_q8_gib": 49.41,
+      "conservative_dsa_q8_gib": 56.98,
+      "conservative_dsa_f16_gib": 107.25
+    }
+  },
+  "aggregate_hard_fit_floor_gib": 224.0,
+  "aggregate_floor_class": "experimental_hard_fit_floor",
+  "placement_imbalance_factor": 1.1,
+  "tier_table": {
+    "32": {
+      "physical_usable_gib": 32.0,
+      "reserve_gib": 8.0,
+      "placement_budget_gib": 24.0,
+      "weight_gib": 201.832,
+      "kv_gib": 0.89,
+      "total_placement_gib": 202.722,
+      "arithmetic_minimum_nodes": 9,
+      "recommended_nodes": 10,
+      "imbalance_factor": 1.1
+    },
+    "48": {
+      "physical_usable_gib": 48.0,
+      "reserve_gib": 9.6,
+      "placement_budget_gib": 38.4,
+      "weight_gib": 201.832,
+      "kv_gib": 0.89,
+      "total_placement_gib": 202.722,
+      "arithmetic_minimum_nodes": 6,
+      "recommended_nodes": 6,
+      "imbalance_factor": 1.1
+    },
+    "64": {
+      "physical_usable_gib": 64.0,
+      "reserve_gib": 12.8,
+      "placement_budget_gib": 51.2,
+      "weight_gib": 201.832,
+      "kv_gib": 0.89,
+      "total_placement_gib": 202.722,
+      "arithmetic_minimum_nodes": 4,
+      "recommended_nodes": 5,
+      "imbalance_factor": 1.1
+    },
+    "96": {
+      "physical_usable_gib": 96.0,
+      "reserve_gib": 19.2,
+      "placement_budget_gib": 76.8,
+      "weight_gib": 201.832,
+      "kv_gib": 0.89,
+      "total_placement_gib": 202.722,
+      "arithmetic_minimum_nodes": 3,
+      "recommended_nodes": 3,
+      "imbalance_factor": 1.1
+    },
+    "128": {
+      "physical_usable_gib": 128.0,
+      "reserve_gib": 25.6,
+      "placement_budget_gib": 102.4,
+      "weight_gib": 201.832,
+      "kv_gib": 0.89,
+      "total_placement_gib": 202.722,
+      "arithmetic_minimum_nodes": 2,
+      "recommended_nodes": 3,
+      "imbalance_factor": 1.1
+    }
+  },
+  "routes": {
+    "recommended_5x64_unified": {
+      "node_count": 5,
+      "aggregate_usable_gib": 320.0,
+      "aggregate_placement_budget_gib": 256.0,
+      "required_placement_gib": 202.722,
+      "fits": true,
+      "meets_hard_fit_floor": true,
+      "no_single_node_can_admit_target": true,
+      "headroom_gib": 53.278,
+      "reasons": []
+    },
+    "recommended_3x96_unified": {
+      "node_count": 3,
+      "aggregate_usable_gib": 288.0,
+      "aggregate_placement_budget_gib": 230.4,
+      "required_placement_gib": 202.722,
+      "fits": true,
+      "meets_hard_fit_floor": true,
+      "no_single_node_can_admit_target": true,
+      "headroom_gib": 27.678,
+      "reasons": []
+    },
+    "recommended_3x128_unified": {
+      "node_count": 3,
+      "aggregate_usable_gib": 384.0,
+      "aggregate_placement_budget_gib": 307.2,
+      "required_placement_gib": 202.722,
+      "fits": true,
+      "meets_hard_fit_floor": true,
+      "no_single_node_can_admit_target": true,
+      "headroom_gib": 104.478,
+      "reasons": []
+    },
+    "fit_probe_2x128_unified": {
+      "node_count": 2,
+      "aggregate_usable_gib": 256.0,
+      "aggregate_placement_budget_gib": 204.8,
+      "required_placement_gib": 202.722,
+      "fits": true,
+      "meets_hard_fit_floor": true,
+      "no_single_node_can_admit_target": true,
+      "headroom_gib": 2.078,
+      "reasons": []
+    },
+    "fit_probe_4x64_unified": {
+      "node_count": 4,
+      "aggregate_usable_gib": 256.0,
+      "aggregate_placement_budget_gib": 204.8,
+      "required_placement_gib": 202.722,
+      "fits": true,
+      "meets_hard_fit_floor": true,
+      "no_single_node_can_admit_target": true,
+      "headroom_gib": 2.078,
+      "reasons": []
+    },
+    "hard_fit_floor_2x112_unified": {
+      "node_count": 2,
+      "aggregate_usable_gib": 224.0,
+      "aggregate_placement_budget_gib": 179.2,
+      "required_placement_gib": 202.722,
+      "fits": false,
+      "meets_hard_fit_floor": true,
+      "no_single_node_can_admit_target": true,
+      "headroom_gib": -23.522,
+      "reasons": [
+        "aggregate placement budget 179.2 GiB is below the 202.7 GiB the target needs after each node's reserve"
+      ]
+    },
+    "insufficient_3x64_unified": {
+      "node_count": 3,
+      "aggregate_usable_gib": 192.0,
+      "aggregate_placement_budget_gib": 153.6,
+      "required_placement_gib": 202.722,
+      "fits": false,
+      "meets_hard_fit_floor": false,
+      "no_single_node_can_admit_target": true,
+      "headroom_gib": -49.122,
+      "reasons": [
+        "aggregate placement budget 153.6 GiB is below the 202.7 GiB the target needs after each node's reserve",
+        "aggregate usable memory 192.0 GiB is below the 224 GiB experimental hard-fit floor"
+      ]
+    }
+  },
+  "seams": {
+    "3_nodes_2.5gbe": {
+      "node_count": 3,
+      "seam_count": 2,
+      "hidden_size": 6144,
+      "bytes_per_token_per_seam": 12288,
+      "prefill_bytes_per_seam": 201326592,
+      "decode_bytes_per_seam_per_token": 12288,
+      "dsa_sideband_bytes_per_query": 8192,
+      "link_rate_gbps": 2.5,
+      "meets_alpha_minimum": true,
+      "is_recommended_link": false,
+      "decode_serialization_ms_per_token": 0.0786,
+      "decode_latency_ms_per_token": 1.0,
+      "decode_bandwidth_share_ms_per_token": 0.0786,
+      "prefill_serialization_ms": 1288.49
+    },
+    "3_nodes_10.0gbe": {
+      "node_count": 3,
+      "seam_count": 2,
+      "hidden_size": 6144,
+      "bytes_per_token_per_seam": 12288,
+      "prefill_bytes_per_seam": 201326592,
+      "decode_bytes_per_seam_per_token": 12288,
+      "dsa_sideband_bytes_per_query": 8192,
+      "link_rate_gbps": 10.0,
+      "meets_alpha_minimum": true,
+      "is_recommended_link": true,
+      "decode_serialization_ms_per_token": 0.0197,
+      "decode_latency_ms_per_token": 1.0,
+      "decode_bandwidth_share_ms_per_token": 0.0197,
+      "prefill_serialization_ms": 322.123
+    },
+    "5_nodes_2.5gbe": {
+      "node_count": 5,
+      "seam_count": 4,
+      "hidden_size": 6144,
+      "bytes_per_token_per_seam": 12288,
+      "prefill_bytes_per_seam": 201326592,
+      "decode_bytes_per_seam_per_token": 12288,
+      "dsa_sideband_bytes_per_query": 8192,
+      "link_rate_gbps": 2.5,
+      "meets_alpha_minimum": true,
+      "is_recommended_link": false,
+      "decode_serialization_ms_per_token": 0.1573,
+      "decode_latency_ms_per_token": 2.0,
+      "decode_bandwidth_share_ms_per_token": 0.1573,
+      "prefill_serialization_ms": 2576.98
+    },
+    "5_nodes_10.0gbe": {
+      "node_count": 5,
+      "seam_count": 4,
+      "hidden_size": 6144,
+      "bytes_per_token_per_seam": 12288,
+      "prefill_bytes_per_seam": 201326592,
+      "decode_bytes_per_seam_per_token": 12288,
+      "dsa_sideband_bytes_per_query": 8192,
+      "link_rate_gbps": 10.0,
+      "meets_alpha_minimum": true,
+      "is_recommended_link": true,
+      "decode_serialization_ms_per_token": 0.0393,
+      "decode_latency_ms_per_token": 2.0,
+      "decode_bandwidth_share_ms_per_token": 0.0393,
+      "prefill_serialization_ms": 644.245
+    }
+  }
+}

.scratch/distributed-gguf-runtime/evidence/DGR-017/upstream-status.json

@@ -0,0 +1,89 @@
+{
+  "observed_at": "2026-07-13",
+  "observed_by": "DGR-017",
+  "method": "GitHub REST API (api.github.com), read-only; no fork, no clone, no patch adopted",
+  "refresh_note": "The roadmap's 2026-07-13 observations were re-verified against live upstream. One item changed: PR #24231 is now MERGED (2026-07-11), which the roadmap already anticipated as 'generic CPU lightning-indexer support is merged'.",
+  "llama_cpp": {
+    "repo": "ggml-org/llama.cpp",
+    "items": [
+      {
+        "ref": "issue #24730",
+        "url": "https://github.com/ggml-org/llama.cpp/issues/24730",
+        "title": "Feature Request: Support for GLM 5.2",
+        "type": "issue",
+        "state": "open",
+        "updated_at": "2026-07-03T22:02:15Z",
+        "meaning": "The umbrella GLM-5.2 support request is still open. GLM-5.2 is not fully supported upstream."
+      },
+      {
+        "ref": "PR #24770",
+        "url": "https://github.com/ggml-org/llama.cpp/pull/24770",
+        "title": "model : glm-dsa load DSA indexer tensors as optional",
+        "type": "pull_request",
+        "state": "closed",
+        "merged_at": "2026-06-20T10:48:24Z",
+        "meaning": "MERGED. GLM-5.2 loads, but through a dense-MLA compatibility path with DSA indexer tensors treated as optional. This is the fallback the alpha contract explicitly refuses: it can produce text without performing DSA/IndexShare computation."
+      },
+      {
+        "ref": "PR #24231",
+        "url": "https://github.com/ggml-org/llama.cpp/pull/24231",
+        "title": "New GGML_OP_LIGHTNING_INDEXER that implements DeepSeek V3.2/V4 lightning indexer",
+        "type": "pull_request",
+        "state": "closed",
+        "merged_at": "2026-07-11T09:39:07Z",
+        "meaning": "MERGED since the roadmap was written. A generic lightning-indexer op now exists in GGML. Backend coverage beyond CPU remains uneven and must be verified per backend by DGR-018, not assumed."
+      },
+      {
+        "ref": "PR #25407",
+        "url": "https://github.com/ggml-org/llama.cpp/pull/25407",
+        "title": "GLM 5.2 Indexer support",
+        "type": "pull_request",
+        "state": "open",
+        "draft": false,
+        "mergeable_state": "unstable",
+        "head_sha": "8dedd06415f36f10fc6091241a39b23c1bf0ee11",
+        "base": "master",
+        "commits": 6,
+        "changed_files": 12,
+        "additions": 414,
+        "deletions": 7,
+        "updated_at": "2026-07-13T15:28:51Z",
+        "meaning": "OPEN and actively moving (updated today). This is the real DSA/IndexShare implementation alpha needs. It is narrow — 12 files, +414/-7 — which is the single most important finding for donor policy: the semantics alpha requires are reviewable and trackable upstream, not a 261-patch fork."
+      }
+    ]
+  },
+  "capability_status_for_alpha": {
+    "gguf_load_of_UD-IQ1_S": "expected via merged #24770, unverified by this project; DGR-018 must prove it against the exact pinned artifact",
+    "moe_routing_and_shared_expert": "expected supported; unverified here",
+    "compressed_mla_kv": "supported via the dense-MLA compatibility path",
+    "dsa_lightning_indexer": "generic GGML op merged (#24231); GLM-5.2 wiring still open (#25407)",
+    "indexshare_full_shared_roles": "NOT upstream; only in open PR #25407",
+    "mtp_nextn": "not required for alpha; NextN tensors must be explicitly loaded or excluded, never silently reinterpreted",
+    "conclusion": "As of 2026-07-13 no released upstream llama.cpp performs native GLM-5.2 DSA + IndexShare. A stock pin today would satisfy 'it emits text' via the dense fallback and would FAIL the alpha semantic-correctness contract. This is the gating technical risk for DGR-004 and DGR-018."
+  },
+  "donor": {
+    "repo": "Mesh-LLM/mesh-llm",
+    "url": "https://github.com/Mesh-LLM/mesh-llm",
+    "license": "Apache-2.0",
+    "stars_observed": 2048,
+    "pushed_at": "2026-07-13T06:45:51Z",
+    "glm_branch": "feat/jianyang-glm-52",
+    "glm_branch_head": "9bd18f1509dff7fac21578635084035b3ba90a38",
+    "glm_branch_head_date": "2026-07-12T06:37:43Z",
+    "policy": "TEST AND PATCH DONOR ONLY. Do not adopt the fork, its scheduler, discovery, routing, public mesh, or package manager. Meshnet remains the sole control plane (RALPH-CONTEXT runtime decision, ADR-0020).",
+    "focused_candidates": [
+      "GLM DSA graph semantics",
+      "lightning indexer and sparse-attention tests",
+      "IndexShare metadata and Full/Shared role validation",
+      "top-k sideband shape and lifecycle",
+      "stage-local KV filtering",
+      "target parity and performance fixtures"
+    ],
+    "adoption_state": "none adopted in DGR-017. This story reads and records upstream state; it takes no patch and forks nothing."
+  },
+  "recommendation_for_dgr_004_and_dgr_018": [
+    "Track upstream PR #25407 rather than forking Mesh-LLM. At 12 files and +414/-7 it is small enough to review, reproduce, and carry as a numbered patch in the project's own pinned stack.",
+    "Any pin chosen before #25407 merges will load GLM-5.2 through the dense-MLA compatibility path. DGR-018 must therefore prove DSA/IndexShare are ACTIVE, not merely that the model emits text — the alpha contract already forbids the fallback.",
+    "Verify lightning-indexer backend coverage (#24231) on the specific backend the route will use. CPU support being merged says nothing about ROCm/HIP."
+  ]
+}

.scratch/distributed-gguf-runtime/evidence/DGR-018/BLOCKED.md

@@ -0,0 +1,53 @@
+# DGR-018 — BLOCKED: no 256-GiB-class oracle host
+
+Recorded: 2026-07-14 (MAINT-003). Preflight scripts preserved at commit
+`a0f28b5` ("chore: preserve DGR-018 preflight scripts (postponed)").
+
+## Blocker
+
+DGR-018 requires a 256-GiB-class host with at least **224 GiB
+runtime-accessible memory** (the DGR-017 experimental hard-fit floor for the
+whole-model `UD-IQ1_S` oracle) and **250 GB free storage** on one filesystem
+outside `/home` (216.715 GB artifact plus resume/temp headroom). The available
+development host fails both gates, so the whole-model oracle cannot be
+established. Per the issue's finish contract, no smaller model may be
+substituted.
+
+DGR-019 (needs the DGR-018 oracle for parity certification) and DGR-020
+(needs DGR-018 and DGR-019, plus enough physical consumer nodes that no single
+node admits the whole recipe) are blocked transitively.
+
+## Exact preflight output
+
+Command (offline; resolves everything from the pinned target manifest and
+never contacts the network):
+
+```
+$ python scripts/glm_whole_model_preflight.py
+target: UD-IQ1_S 216.715 GB, 6 shards @ abc55e725277
+[FAIL] storage: need >= 250 GB free on one filesystem outside ['/home']; observed no eligible filesystem
+[FAIL] memory: need >= 224 GiB runtime-accessible memory (DGR-017 experimental hard-fit floor); observed 124.9 GiB MemTotal
+destination: NONE — no filesystem outside ['/home'] has 250 GB free
+  - /run/media/popov/DATA (ext4): 74.2 GB free
+  - / (ext4): 51.1 GB free
+  - /run/media/popov/Windows (fuseblk): 26.0 GB free
+  - /run/media/popov/d (fuseblk): 5.1 GB free
+verdict: fail
+$ echo $?
+1
+```
+
+Host: Linux 7.0.14-101.fc43.x86_64 x86_64, `MemTotal: 130997376 kB`
+(124.9 GiB). The full machine-readable report (including the ordered
+download/verify plan against revision `abc55e72527792c6e77069c99b4cb7de16fa9f23`,
+manifest SHA-256 `0b6aed04479d204902bb64c0203f1a46cab26a47b378ecccf85237b63f6c1962`)
+is in [preflight.json](preflight.json).
+
+## How to resume
+
+1. On a qualifying host, run `python scripts/glm_whole_model_preflight.py`
+   (optionally `--dest DIR`); it must exit 0 with `verdict: pass`.
+2. Download shards in the preflight's ordered plan; verify each with
+   `python scripts/verify_glm_shards.py` before the next transfer starts.
+3. Proceed with the DGR-018 issue
+   (`.scratch/distributed-gguf-runtime/issues/18-certify-whole-model-glm-5-2-runtime-semantics.md`).

.scratch/distributed-gguf-runtime/evidence/DGR-018/preflight.json

@@ -0,0 +1,140 @@
+{
+  "generated_by": "scripts/glm_whole_model_preflight.py",
+  "target": {
+    "gguf_repo_id": "unsloth/GLM-5.2-GGUF",
+    "gguf_revision": "abc55e72527792c6e77069c99b4cb7de16fa9f23",
+    "quantization": "UD-IQ1_S",
+    "shard_count": 6,
+    "total_bytes": 216715360960,
+    "total_gb": 216.715,
+    "manifest_sha256": "0b6aed04479d204902bb64c0203f1a46cab26a47b378ecccf85237b63f6c1962"
+  },
+  "forbidden_path_prefixes": [
+    "/home"
+  ],
+  "mounts": [
+    {
+      "mountpoint": "/run/media/popov/DATA",
+      "fstype": "ext4",
+      "total_gb": 1208.8,
+      "free_gb": 74.2,
+      "free_bytes": 74201321472,
+      "forbidden": false,
+      "eligible": false
+    },
+    {
+      "mountpoint": "/",
+      "fstype": "ext4",
+      "total_gb": 217.7,
+      "free_gb": 51.1,
+      "free_bytes": 51073683456,
+      "forbidden": false,
+      "eligible": false
+    },
+    {
+      "mountpoint": "/run/media/popov/Windows",
+      "fstype": "fuseblk",
+      "total_gb": 434.9,
+      "free_gb": 26.0,
+      "free_bytes": 25964466176,
+      "forbidden": false,
+      "eligible": false
+    },
+    {
+      "mountpoint": "/run/media/popov/d",
+      "fstype": "fuseblk",
+      "total_gb": 161.1,
+      "free_gb": 5.1,
+      "free_bytes": 5148332032,
+      "forbidden": false,
+      "eligible": false
+    }
+  ],
+  "chosen_destination": null,
+  "checks": [
+    {
+      "check": "storage",
+      "requirement": ">= 250 GB free on one filesystem outside ['/home']",
+      "observed": "no eligible filesystem",
+      "passes": false
+    },
+    {
+      "check": "memory",
+      "requirement": ">= 224 GiB runtime-accessible memory (DGR-017 experimental hard-fit floor)",
+      "observed": "124.9 GiB MemTotal",
+      "passes": false,
+      "waived": false
+    }
+  ],
+  "download_authorized": false,
+  "storage_only": false,
+  "download_plan": [
+    {
+      "step": 1,
+      "shard_index": 1,
+      "path": "UD-IQ1_S/GLM-5.2-UD-IQ1_S-00001-of-00006.gguf",
+      "size_bytes": 9423744,
+      "size_gb": 0.009,
+      "sha256": "46b6148389219ae45167cb8124fbb18ef7d432daf619b4faf9e06ea80d3f4777",
+      "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00001-of-00006.gguf",
+      "download_command": "curl -L -C - --fail -o \"$GLM_DEST/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00001-of-00006.gguf\" \"https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00001-of-00006.gguf\"",
+      "verify_command": "python scripts/verify_glm_shards.py --model-dir \"$GLM_DEST\" --shard 1"
+    },
+    {
+      "step": 2,
+      "shard_index": 6,
+      "path": "UD-IQ1_S/GLM-5.2-UD-IQ1_S-00006-of-00006.gguf",
+      "size_bytes": 19171063136,
+      "size_gb": 19.171,
+      "sha256": "3b767f55df64e0432d52fcf1a14eb47a1ef3bbc91339e2ae220f38602237d7d7",
+      "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00006-of-00006.gguf",
+      "download_command": "curl -L -C - --fail -o \"$GLM_DEST/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00006-of-00006.gguf\" \"https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00006-of-00006.gguf\"",
+      "verify_command": "python scripts/verify_glm_shards.py --model-dir \"$GLM_DEST\" --shard 6"
+    },
+    {
+      "step": 3,
+      "shard_index": 2,
+      "path": "UD-IQ1_S/GLM-5.2-UD-IQ1_S-00002-of-00006.gguf",
+      "size_bytes": 49208128256,
+      "size_gb": 49.208,
+      "sha256": "f2180207285e04fcaa5b8c53ba6e77ad5cc58666b6e7c6b04a5eded3fe8bef09",
+      "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00002-of-00006.gguf",
+      "download_command": "curl -L -C - --fail -o \"$GLM_DEST/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00002-of-00006.gguf\" \"https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00002-of-00006.gguf\"",
+      "verify_command": "python scripts/verify_glm_shards.py --model-dir \"$GLM_DEST\" --shard 2"
+    },
+    {
+      "step": 4,
+      "shard_index": 3,
+      "path": "UD-IQ1_S/GLM-5.2-UD-IQ1_S-00003-of-00006.gguf",
+      "size_bytes": 49684417024,
+      "size_gb": 49.684,
+      "sha256": "b1c0c5a302cc8d5d9ea0bcd4467c01db72c26839f820f7e882079582ea0a8d2b",
+      "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00003-of-00006.gguf",
+      "download_command": "curl -L -C - --fail -o \"$GLM_DEST/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00003-of-00006.gguf\" \"https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00003-of-00006.gguf\"",
+      "verify_command": "python scripts/verify_glm_shards.py --model-dir \"$GLM_DEST\" --shard 3"
+    },
+    {
+      "step": 5,
+      "shard_index": 4,
+      "path": "UD-IQ1_S/GLM-5.2-UD-IQ1_S-00004-of-00006.gguf",
+      "size_bytes": 49396052864,
+      "size_gb": 49.396,
+      "sha256": "a6a42da6975e29f89866dcde2956e9e50e6ea26635fb5063b74f3973f4f863b6",
+      "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00004-of-00006.gguf",
+      "download_command": "curl -L -C - --fail -o \"$GLM_DEST/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00004-of-00006.gguf\" \"https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00004-of-00006.gguf\"",
+      "verify_command": "python scripts/verify_glm_shards.py --model-dir \"$GLM_DEST\" --shard 4"
+    },
+    {
+      "step": 6,
+      "shard_index": 5,
+      "path": "UD-IQ1_S/GLM-5.2-UD-IQ1_S-00005-of-00006.gguf",
+      "size_bytes": 49246275936,
+      "size_gb": 49.246,
+      "sha256": "a4a9851a50db533f21ef824e5d8038f04e6782e7d602d18e5fdd6643f68ccccb",
+      "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00005-of-00006.gguf",
+      "download_command": "curl -L -C - --fail -o \"$GLM_DEST/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00005-of-00006.gguf\" \"https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00005-of-00006.gguf\"",
+      "verify_command": "python scripts/verify_glm_shards.py --model-dir \"$GLM_DEST\" --shard 5"
+    }
+  ],
+  "verdict": "fail"
+}

.scratch/distributed-gguf-runtime/evidence/MAINTENANCE-HANDOFF.md

@@ -0,0 +1,62 @@
+# Maintenance review handoff — distributed GGUF runtime
+
+Date: 2026-07-14
+Scope: close the maintenance review, preserve the hard blockers, and hand off the remaining implementation work to the next model.
+
+## What is complete
+
+- Completed stories are now recorded in `docs/issues/distributed-gguf-runtime/`.
+- The PRD and milestone docs were updated to reflect the closed set and the blocked set.
+- The DGR-018 preflight scripts were preserved at commit `a0f28b5`.
+- The current feature line has delivered DGR-001 through DGR-006 and DGR-017.
+
+## Hard / unsolved issues for later
+
+### 1) DGR-018 requires hardware we do not have
+
+DGR-018 is blocked because the whole-model GLM-5.2 UD-IQ1_S oracle requires:
+
+- a **256-GiB-class host**,
+- at least **224 GiB runtime-accessible memory**,
+- at least **250 GB free storage on one filesystem outside `/home`**.
+
+The current development host reports only **124.9 GiB MemTotal** and has no eligible filesystem with 250 GB free.
+The authoritative blocker evidence is in `evidence/DGR-018/BLOCKED.md` and `evidence/DGR-018/preflight.json`.
+
+### 2) DGR-019 and DGR-020 are transitively blocked
+
+- **DGR-019** needs the DGR-018 oracle for parity certification.
+- **DGR-020** needs DGR-018 and DGR-019, plus enough physical consumer nodes that no single node can admit the whole recipe.
+
+No smaller model may be substituted for these stories.
+
+### 3) The remainder of the graph stays blocked unless replanned
+
+The current graph makes **DGR-007 depend on DGR-019**, which means:
+
+- DGR-007 through DGR-016 are also blocked transitively.
+- Unblocking the dense pipeline without the 256-GiB host would require an explicit replanning decision to relax the DGR-007 → DGR-019 dependency.
+- That replanning decision has **not** been made.
+
+### 4) Maintenance-only tasks should stay separate from feature implementation
+
+The review uncovered that the codebase now has a clean closed-story split, but further work should avoid mixing:
+
+- maintenance cleanup,
+- blocked-hardware preparation,
+- and actual distributed GLM implementation.
+
+The next model should treat the maintenance pass as closed and only pick up real implementation work that is not hardware-blocked.
+
+## Recommended next move
+
+Use the next model to continue on the **non-blocked implementation queue** only.
+Priority candidates are whatever is still actionable without the GLM oracle host; if a story depends on DGR-018, keep it deferred.
+
+## Reference files
+
+- `docs/issues/distributed-gguf-runtime/README.md`
+- `.scratch/distributed-gguf-runtime/PRD.md`
+- `.scratch/distributed-gguf-runtime/milestones.md`
+- `.scratch/distributed-gguf-runtime/evidence/DGR-018/BLOCKED.md`
+- `.scratch/distributed-gguf-runtime/evidence/DGR-018/preflight.json`

.scratch/distributed-gguf-runtime/evidence/README.md

@@ -0,0 +1,15 @@
+# Ralph task evidence
+
+Each completed story creates `evidence/<TASK-ID>/README.md`. Fresh dependent iterations must read it before coding.
+
+Required README sections:
+
+1. Summary and acceptance decision.
+2. Exact files changed.
+3. Commands run and real exit/results.
+4. Correctness, performance and hardware evidence classification.
+5. Known limitations and deferred work.
+6. Compatibility/migration notes.
+7. Explicit handoff for each dependent story.
+
+Store raw machine-readable metrics, manifests and protocol artifacts beside the README. Never store secrets, model weights, build outputs or Ralph iteration logs here.

.scratch/distributed-gguf-runtime/implementation-strategy.md

@@ -0,0 +1,247 @@
+# Focused implementation strategy: performant concurrent distributed inference
+
+Status: Accepted planning direction
+Last updated: 2026-07-13
+
+## Product objective
+
+Enable clients to run top open models that do not fit on one consumer machine by combining independently owned model Shards into performant, concurrent Inference Routes.
+
+The alpha-release target is the exact `zai-org/GLM-5.2` model, pinned by revision and served with `reasoning_effort=max`, using the smallest published Unsloth `UD-IQ1_S` GGUF across physical consumer machines. See [GLM-5.2-MAX-ALPHA-ROADMAP.md](GLM-5.2-MAX-ALPHA-ROADMAP.md). Dense Llama remains a cheap structural fixture; Qwen expansion is post-alpha.
+
+The project is not trying to reproduce every vLLM feature or support every inference engine. It is optimizing for:
+
+1. Models larger than one node's RAM/VRAM.
+2. Useful interactive decode speed on consumer CPU, AMD, NVIDIA, Vulkan, and mixed routes where certified.
+3. Multiple concurrent Route Sessions without cache corruption or global serialization.
+4. A lean runtime with one control plane and one primary GGUF engine.
+5. Measured improvement over the existing Transformers/safetensors implementation.
+
+## Current reality
+
+The existing project already owns the differentiating distributed control plane:
+
+- Tracker-selected contiguous Shards.
+- Stable Route Sessions.
+- Local per-Shard Hot KV State in the Transformers reference backend.
+- Binary Activation Seams.
+- Relay/direct routing, cancellation, telemetry, billing, and capability admission.
+- Persistent relay and direct transport optimizations.
+
+The missing production path is a native GGUF execution worker that can load and execute only an assigned layer range while retaining local Hot KV State for concurrent Route Sessions.
+
+Whole-model llama.cpp, vLLM, and existing Transformers serving remain baselines or optional route kinds. They are not substitutes for native distributed Shards.
+
+## Performance hypothesis—not an assumption
+
+GGUF itself is a format. Performance comes from llama.cpp/GGML's quantized kernels, memory layout, mmap, backend scheduling, and reduced working set.
+
+Quantized GGUF may be faster or may merely fit a larger model. Comparisons against safetensors must report both speed and quality because BF16 safetensors and Q4/Q8 GGUF are not numerically equivalent.
+
+Before expensive native work, establish controlled lanes. DGR-001 remains immutable; DGR-017 adds a target-specific fit and semantics contract without rewriting DGR-001 evidence:
+
+- Same model architecture and upstream revision.
+- Same machine, prompt set, context, output length, sampling policy, and concurrency.
+- Transformers/safetensors BF16 or the current production recipe.
+- llama.cpp GGUF F16/BF16 or Q8 correctness lane where available.
+- Q4_K_M or selected production quantization performance/fit lane.
+- TTFT, prefill tok/s, decode tok/s, p50/p95 latency, RSS, VRAM, artifact size, energy where available, and output-quality drift.
+
+The program proceeds only if llama.cpp/GGUF provides at least one meaningful advantage recorded in a machine-readable performance contract:
+
+- Better decode or aggregate throughput at acceptable quality; or
+- Materially lower memory that makes the target model routable while preserving useful throughput.
+
+## Parallelism we will use
+
+### Public Inference Route: layer/pipeline parallelism
+
+Each node independently executes one contiguous Shard. Activations cross seams; weights and Hot KV State remain local.
+
+This is the only public cross-machine model-parallel primitive in the first runtime.
+
+### Per-node continuous batching
+
+Autoregressive tokens remain sequential within one generation. Throughput comes from batching decode steps from multiple active Route Sessions inside each node using llama.cpp batches and sequence IDs or bounded context pools.
+
+This is essential. A worker that globally serializes sessions is not production-ready.
+
+### Multiple complete routes: data parallelism
+
+The Tracker may select multiple complete routes for independent requests. This increases network throughput and availability without requiring collectives between routes.
+
+### Trusted composite node: optional tensor/expert parallelism
+
+Tensor parallelism and expert parallelism require frequent collectives and tight compatibility. They may be used later inside one operator-controlled composite node or managed cluster exposed as one logical provider. They are not public WAN routing primitives.
+
+### Deferred mechanisms
+
+- Disaggregated prefill and KV transfer.
+- Speculative decoding.
+- Cross-route prefix snapshots.
+- Route repair with KV migration.
+- Public tensor/expert parallel collectives.
+
+They remain out of the critical path until the native layer route passes performance and concurrency gates.
+
+## Reuse decisions
+
+### llama.cpp/GGML: primary runtime substrate
+
+Reuse:
+
+- GGUF parsing and mmap.
+- Quantized kernels.
+- CPU, CUDA, HIP/ROCm, Vulkan, Metal, and other supported backends.
+- Tokenizer and model architecture implementations.
+- KV and sequence operations.
+- Backend scheduler and graph execution.
+
+Maintain a small exact-commit fork only for the missing local seam:
+
+- Range-aware tensor ownership/loading.
+- Architecture-defined boundary input/output.
+- Intermediate boundary output without tail normalization.
+- Layer-filtered KV and sequence mapping.
+
+Keep networking, Tracker logic, billing, and public protocol outside llama.cpp. Upstream generic hooks where possible.
+
+### vLLM: concepts and optional managed backend
+
+Use unmodified vLLM only as:
+
+- A whole-model node backend.
+- A managed TP/PP/EP cluster represented as one logical provider.
+- A performance/correctness baseline.
+
+Adapt concepts, not runtime code:
+
+- Named intermediate tensor bundles.
+- Continuous batching and request-owner maps.
+- Versioned KV-transfer compatibility fingerprints.
+- Explicit send/receive/abort/failure lifecycle.
+- Load telemetry and unbiased route selection.
+
+Do not fork vLLM for public Shards and do not transplant PagedAttention, Torch process groups, or GGUF-plugin kernels into the llama.cpp worker.
+
+### Nakshatra, prima.cpp, llama-gguf, LiGGUF, GPUStack
+
+Use as source and test donors only:
+
+- Nakshatra: partial-GGUF patches, daemon concepts, replay cases.
+- prima.cpp: selected tensor ownership and local-layer KV evidence.
+- llama-gguf: small protocol and integration-test patterns.
+- LiGGUF: Q8 activation transport and tensor-reduction reference.
+- historical GPUStack: resource preflight and role-oriented placement.
+
+Do not adopt or fork their repositories wholesale.
+
+### Mesh-LLM GLM branch: focused test/patch donor only
+
+Use its GLM-5.2 branch to study DSA, IndexShare, stage-local KV, and sideband tests. Do not import its scheduler, discovery/control plane, package manager, or broad llama.cpp patch stack. Every adopted idea must be independently understood, minimized, attributed, and tested against our exact pin.
+
+## Battle-proven transport decision
+
+Use gRPC over HTTP/2 with Protocol Buffers for the native C++ Shard worker protocol.
+
+Why:
+
+- Mature Python and C++ implementations.
+- Bidirectional streaming.
+- HTTP/2 flow control and connection reuse.
+- Deadlines, cancellation, status codes, TLS, authentication interceptors, and generated schemas.
+- Avoids inventing a socket protocol.
+
+Scope boundary:
+
+- OpenAI-compatible client/Gateway APIs remain HTTP/SSE.
+- Tracker/control APIs remain existing project interfaces.
+- One long-lived bidirectional gRPC stream serves one Route Session Activation Seam.
+- Existing relay/WebSocket infrastructure may carry the same versioned protobuf frames as opaque binary when direct gRPC reachability is unavailable.
+- Large prefill tensors are chunked into bounded frames; decode bundles stay small.
+- No QUIC/WebRTC/custom transport in this milestone.
+
+The public boundary uses a versioned named-tensor bundle rather than one anonymous tensor because architecture boundaries can require more than `hidden_states`. DGR-006 updates the current single-`NamedTensor` decode fast path to carry the same bundle semantics and adds an explicit typed tail logits/token result with sampling/template identity.
+
+Minimum identity:
+
+```text
+schema version
+request/work id
+Route Session id and route epoch
+Model Artifact and runtime recipe fingerprint
+Shard range and effective start
+phase: prefill/decode/release/cancel
+position/token range
+named tensors with shape/dtype/byte order
+compression and checksum
+idempotency step id
+cache expectation/result
+```
+
+## Concurrency model
+
+A native worker must not use one global serving sequence or one lock around all model execution.
+
+Required ownership:
+
+```text
+(Route Session id, route epoch)
+    -> local sequence/context
+    -> Shard-local Hot KV State
+    -> bounded lease and memory accounting
+```
+
+The node scheduler:
+
+- Admits sessions against model memory and KV budget.
+- Forms compatible decode batches from active sessions.
+- Preserves per-session position and route order.
+- Applies bounded queues and backpressure.
+- Cancels/releases independently.
+- Reports queue, batch, KV, prefill, decode, and seam telemetry.
+
+Initial deterministic gate: at least four concurrent sessions on a small certified model with no token/KV cross-talk. Final concurrency targets are hardware/recipe-specific and recorded by capability admission rather than hardcoded globally.
+
+## Stage gates
+
+### Gate A: performance hypothesis
+
+Controlled safetensors-versus-GGUF benchmark produces a signed/reproducible report and locks thresholds. Stop native work if there is no meaningful speed or fit benefit.
+
+### Gate B: local range parity
+
+Two local processes own disjoint GGUF ranges and match whole-model llama.cpp within the certified numerical tolerance for prefill and greedy decode.
+
+### Gate C: concurrent KV
+
+Multiple Route Sessions prefill/decode concurrently with isolated local KV, bounded memory, cancellation, and release.
+
+### Gate D: real distributed route
+
+Two physical machines execute one model that uses both Shards. Synthetic activation tests do not satisfy this gate.
+
+### Gate E: consumer-hardware performance
+
+On certified consumer hardware, the GGUF route beats the current distributed safetensors route under the locked performance contract or enables a larger otherwise-unroutable model at useful measured speed.
+
+### Gate F: exact GLM-5.2 alpha target
+
+After the generic dense fixture proves range and boundary mechanics, certify explicit GLM-5.2 MoE, MLA KV, DSA, IndexShare, and NextN policy. Alpha requires the exact `UD-IQ1_S` target across physical consumer nodes, native Max-mode semantics, locked parity/usefulness/performance thresholds, and bounded failure cleanup. Qwen3/Qwen3-MoE is later architecture expansion.
+
+## Scope discipline
+
+The following do not block the first production candidate:
+
+- New cryptocurrency/economics work.
+- New artifact P2P protocol.
+- QUIC or WebRTC.
+- vLLM fork.
+- Whole-repository Nakshatra/prima adoption.
+- Every GGUF architecture.
+- Automatic route repair.
+- Prefix snapshot migration.
+- Speculative decoding.
+- A large-model marketing demo before small-model parity and concurrency pass.
+
+Every optimization must preserve output contract, session isolation, cancellation, resource cleanup, capability admission, and per-node attribution.

.scratch/distributed-gguf-runtime/issues/01-route-session-lifecycle.md

@@ -1,19 +0,0 @@
-# 01 — Route Session lifecycle
-
-Status: ready-for-agent
-
-## What to build
-
-Add the narrowest end-to-end Route Session lifecycle that can be used by distributed inference routes: create a session, bind it to a selected Inference Route, expose status, and close it cleanly. This slice does not need real model cache yet; it proves stable session identity across the control plane and activation plane.
-
-## Acceptance criteria
-
-- [ ] A request can create a Route Session with a stable `session_id`, `route_id`, model preset, backend id, and route membership.
-- [ ] Every downstream activation request carries the same session identity and fails clearly if the session or route id does not match.
-- [ ] Session status reports phase, route nodes, model preset, backend id, created time, and last activity time.
-- [ ] Closing a session releases all registered per-session state.
-- [ ] Tests cover create, status, close, stale-session rejection, and wrong-route rejection.
-
-## Blocked by
-
-None - can start immediately.

.scratch/distributed-gguf-runtime/issues/02-prefill-decode-binary-http.md

@@ -1,20 +0,0 @@
-# 02 — Prefill/decode binary HTTP protocol
-
-Status: ready-for-agent
-
-## What to build
-
-Split the activation protocol into explicit prefill and decode-step calls using the existing binary HTTP direction from ADR-0008. The completed slice should work against a stub backend so payload shape, route/session headers, relay preservation, and failure behavior are testable before real KV cache work begins.
-
-## Acceptance criteria
-
-- [ ] Prefill accepts chunked binary activations with route/session metadata and forwards them through the selected route.
-- [ ] Decode-step accepts a one-step binary activation and forwards a one-step activation through the selected route.
-- [ ] Decode-step payload size is independent of prompt length in protocol tests.
-- [ ] Relay forwarding preserves route/session headers, shape, dtype, position, and wire version.
-- [ ] Legacy `/forward` either remains as a compatibility wrapper or fails with a clear wire-version error.
-- [ ] Tests cover prefill chunking, decode-step shape validation, relay preservation, and malformed header rejection.
-
-## Blocked by
-
-- 01 — Route Session lifecycle.

.scratch/distributed-gguf-runtime/issues/03-generation-telemetry-and-streaming.md

@@ -1,21 +0,0 @@
-# 03 — Generation Telemetry and streaming response contract
-
-Status: ready-for-agent
-
-## What to build
-
-Expose realtime Generation Telemetry for active Route Sessions and stream token deltas when the serving path can produce them. This slice should make long distributed requests observable before real large-model work begins.
-
-## Acceptance criteria
-
-- [ ] A client can observe route-session phase changes: queued, loading, prefill, decode, finalizing, completed, failed.
-- [ ] Telemetry includes prefill progress, generated token count, rolling tokens/sec, average tokens/sec, active route nodes, and failure reason.
-- [ ] Telemetry is available before the first output token.
-- [ ] A streaming response can include token deltas while telemetry remains available.
-- [ ] A non-streaming fallback still exposes telemetry until final answer or failure.
-- [ ] Route-node failure reports the last known phase and reason.
-- [ ] Tests cover telemetry updates, streaming token deltas, non-streaming fallback, and structured failure closeout.
-
-## Blocked by
-
-- 01 — Route Session lifecycle.

.scratch/distributed-gguf-runtime/issues/04-pytorch-distributed-kv-reference.md

@@ -1,23 +0,0 @@
-# 04 — PyTorch distributed KV reference route
-
-Status: ready-for-agent
-
-## What to build
-
-Fix the existing distributed PyTorch route so it uses the Route Session and prefill/decode protocol to keep Hot KV State local to each Shard node. The visible behavior is that prefill processes the prompt once, and decode no longer recomputes or resends the full growing prompt for every token.
-
-## Acceptance criteria
-
-- [ ] Distributed PyTorch prefill stores per-session cache/state on each Shard node.
-- [ ] Distributed PyTorch decode-step reads and appends local per-shard cache/state.
-- [ ] Decode activation seam payload is one token/hidden-state step after prefill.
-- [ ] The old full-growing-prompt decode loop is not used for models that support the reference cache path.
-- [ ] Unsupported model/cache APIs fail with an explicit backend capability error.
-- [ ] Session close or TTL cleanup releases per-shard cache.
-- [ ] Regression tests prove decode does not call the full prompt encoder for every generated token.
-
-## Blocked by
-
-- 01 — Route Session lifecycle.
-- 02 — Prefill/decode binary HTTP protocol.
-- 03 — Generation Telemetry and streaming response contract.

.scratch/distributed-gguf-runtime/issues/05-local-llamacpp-gguf-backend.md

@@ -1,20 +0,0 @@
-# 05 — Local llama.cpp/GGUF backend
-
-Status: ready-for-agent
-
-## What to build
-
-Add a local full-model GGUF backend so a node that can hold a GGUF model can serve it through the existing node API. This is the immediate CPU-performance path and the baseline for later distributed llama.cpp work.
-
-## Acceptance criteria
-
-- [ ] A node can start with backend `llama.cpp` or `gguf` for a local full-model GGUF artifact.
-- [ ] The node can answer an OpenAI-compatible chat completion through the existing API.
-- [ ] Startup and registration clearly report backend, quantization/artifact metadata, context cap, and local model path.
-- [ ] The PyTorch backend remains unchanged and selectable.
-- [ ] A smoke test or script validates backend wiring with a small GGUF model or a stubbed llama.cpp process.
-- [ ] A benchmark command can compare local PyTorch CPU and local GGUF CPU for the same small supported model when both are available.
-
-## Blocked by
-
-None - can start immediately.

.scratch/distributed-gguf-runtime/issues/06-model-artifact-manifest.md

@@ -1,20 +0,0 @@
-# 06 — Model Artifact manifest and Shard advertisement
-
-Status: ready-for-agent
-
-## What to build
-
-Introduce a Model Artifact manifest that separates storage distribution from route execution. A node should be able to verify local model files, determine which Shards it can serve, and advertise artifact/layer availability to the Tracker without contacting Hugging Face at request time.
-
-## Acceptance criteria
-
-- [ ] Manifest records model preset, upstream revision, license, backend support, quantization, context cap, tokenizer artifacts, file hashes, piece hashes, and tensor/layer mapping where available.
-- [ ] A node can verify local artifacts against the manifest and reject corrupt or incomplete artifacts.
-- [ ] A node can derive advertised Shard ranges from the manifest and local files.
-- [ ] Tracker registration can include artifact id, backend id, Shard range, and verification status.
-- [ ] Tracker coverage can distinguish model-layer coverage from artifact availability.
-- [ ] Tests cover valid manifest registration, corrupt artifact rejection, and missing layer/tensor metadata.
-
-## Blocked by
-
-- 01 — Route Session lifecycle.

.scratch/distributed-gguf-runtime/issues/07-add-isolated-concurrent-local-hot-kv-state.md

@@ -0,0 +1,60 @@
+# 07 — Add isolated concurrent local Hot KV State
+
+Status: ready-for-agent
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-007` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a client, I need concurrent Route Sessions to retain independent per-Shard cache so that one request cannot clear or corrupt another.
+
+## Expected durable outputs
+
+- Concurrent local KV/session manager
+- Isolation, eviction, cancellation and cleanup tests
+- evidence/DGR-007/README.md
+
+## Acceptance criteria
+
+- [ ] Map `(Route Session ID, route epoch)` to an isolated llama sequence or bounded context.
+- [ ] Allocate KV only for owned layers.
+- [ ] Support prefill append, decode append, truncate, release, TTL/LRU eviction, and explicit cache-miss response.
+- [ ] Reject stale epochs and incompatible cache recipes.
+- [ ] At least four concurrent sessions on a small model complete without token or KV cross-talk.
+- [ ] Cancellation/release of one session leaves other sessions intact and memory returns to the configured budget.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched
+- [ ] llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-007/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-006` and `DGR-019` must have `passes: true`; read both evidence READMEs and verify their referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

.scratch/distributed-gguf-runtime/issues/07-llamacpp-layer-boundary-prototype.md

@@ -1,25 +0,0 @@
-# 07 — llama.cpp layer-boundary prototype
-
-Status: ready-for-human
-
-## What to build
-
-Build a local prototype that proves whether llama.cpp/libllama can support the platform's distributed execution contract: execute a selected layer range, accept inbound hidden states, emit outbound hidden states, and own per-session cache for only the loaded/served range.
-
-This is the collaboration package for upstream llama.cpp. The target is an upstreamable API shape, not a permanent fork.
-
-## Acceptance criteria
-
-- [ ] A small llama.cpp-supported GGUF model can be split into a two-process localhost head/tail prototype.
-- [ ] The head process runs embeddings and early layers, then emits hidden states at an Activation Seam.
-- [ ] The tail process accepts hidden states, runs later layers plus output head, and produces logits/tokens comparable to single-process execution.
-- [ ] Prefill is performed once and decode-step seam payload is one hidden-state step per generated token.
-- [ ] Each process owns only its own per-session cache/state.
-- [ ] The prototype records the minimum upstream API needed for layer-range execution, hidden-state I/O, partial loading/introspection, and per-session KV ownership.
-- [ ] If upstream support is unavailable, the issue ends with a concrete recommendation: upstream proposal, narrow adapter fork, or keep GGUF distribution local-only for now.
-
-## Blocked by
-
-- 02 — Prefill/decode binary HTTP protocol.
-- 05 — Local llama.cpp/GGUF backend.
-- 06 — Model Artifact manifest and Shard advertisement.

.scratch/distributed-gguf-runtime/issues/08-build-the-standalone-c-grpc-shard-worker.md

@@ -0,0 +1,65 @@
+# 08 — Build the standalone C++ gRPC Shard worker
+
+Status: ready-for-agent
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-008` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a node runtime, I need one supervised native process so that llama.cpp internals remain behind a stable project-owned protocol.
+
+## Expected durable outputs
+
+- Standalone C++ gRPC worker
+- Fake-model Python/C++ integration tests
+- Lifecycle and bounded-failure evidence
+- evidence/DGR-008/README.md
+
+## Acceptance criteria
+
+- [ ] Worker exposes capability, health, session stream, release, cancellation, and metrics services from DGR-002.
+- [ ] Worker loads one exact Artifact/recipe/Shard identity and refuses mismatched requests.
+- [ ] Streaming path enforces bounded messages, flow control, deadlines, idempotency, and independent session cancellation.
+- [ ] Worker does not expose raw llama.cpp RPC or arbitrary GGML graph execution.
+- [ ] Graceful shutdown releases sessions; crash behavior is bounded and observable.
+- [ ] Python integration tests run against a fake model mode without model downloads.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched
+- [ ] llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-008/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-002` must have `passes: true`; read `../evidence/DGR-002/README.md` and verify its referenced files/commands.
+- `DGR-003` must have `passes: true`; read `../evidence/DGR-003/README.md` and verify its referenced files/commands.
+- `DGR-004` must have `passes: true`; read `../evidence/DGR-004/README.md` and verify its referenced files/commands.
+- `DGR-006` must have `passes: true`; read `../evidence/DGR-006/README.md` and verify its referenced files/commands.
+- `DGR-007` must have `passes: true`; read `../evidence/DGR-007/README.md` and verify its referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

.scratch/distributed-gguf-runtime/issues/08-networked-distributed-gguf-route.md

@@ -1,24 +0,0 @@
-# 08 — Networked distributed GGUF route
-
-Status: pending
-
-## What to build
-
-Run a GGUF-backed model over a real multi-node Inference Route using the resolved Route Session, binary HTTP prefill/decode protocol, local Hot KV State, Generation Telemetry, and alpha fail-fast behavior.
-
-## Acceptance criteria
-
-- [ ] Two machines can form one GGUF-backed Inference Route over contiguous Shards.
-- [ ] Prefill builds local per-shard cache/state and decode-step uses one-step seam payloads.
-- [ ] The client receives streamed token deltas when supported by the GGUF path.
-- [ ] The client receives Generation Telemetry for phase, generated tokens, tokens/sec, route health, and failure reason.
-- [ ] Route-node loss fails the Route Session cleanly; no automatic repair is attempted in alpha.
-- [ ] Tracker metrics show prefill tokens/sec, decode tokens/sec, seam latency, queue depth, and cache memory by node.
-- [ ] Billing/audit records identify route membership and layer/token work for the completed or failed session.
-
-## Blocked by
-
-- 03 — Generation Telemetry and streaming response contract.
-- 04 — PyTorch distributed KV reference route.
-- 06 — Model Artifact manifest and Shard advertisement.
-- 07 — llama.cpp layer-boundary prototype.

.scratch/distributed-gguf-runtime/issues/09-deepseek-v4-flash-support-audit.md

@@ -1,21 +0,0 @@
-# 09 — DeepSeek-V4-Flash support audit
-
-Status: ready-for-agent
-
-## What to build
-
-Audit `deepseek-ai/DeepSeek-V4-Flash` as the first serious large-model target after the small GGUF protocol smoke test. The output is a compatibility matrix and a recommended runtime path, not full production support.
-
-## Acceptance criteria
-
-- [ ] Verify current PyTorch/Transformers load and generation semantics for DeepSeek-V4-Flash from primary model documentation.
-- [ ] Verify vLLM and SGLang support status from primary runtime documentation or release notes.
-- [ ] Verify whether a GGUF/llama.cpp quantization path exists or would need upstream work.
-- [ ] Estimate artifact size, active parameter behavior, and 128K cache memory by Shard range.
-- [ ] Identify required backend capability flags for the Tracker.
-- [ ] Produce a compatibility matrix: PyTorch, vLLM, SGLang, llama.cpp/GGUF.
-- [ ] End with one recommendation: first runtime path, blocked pending upstream, or defer.
-
-## Blocked by
-
-None - can start immediately.

.scratch/distributed-gguf-runtime/issues/09-integrate-the-native-worker-with-meshnet.md

@@ -0,0 +1,61 @@
+# 09 — Integrate the native worker with Meshnet
+
+Status: ready-for-agent
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-009` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As the existing node service, I need a GGUF Shard backend adapter so that the Tracker, relay, billing, telemetry, and capability admission remain the sole control plane.
+
+## Expected durable outputs
+
+- Meshnet GGUF backend adapter
+- Registration, routing, relay, telemetry and billing tests
+- evidence/DGR-009/README.md
+
+## Acceptance criteria
+
+- [ ] Implement the existing model-backend surface without changing Transformers behavior.
+- [ ] Registration carries exact validated GGUF recipe, Shard, backend and concurrency/KV capacity.
+- [ ] Tracker forms only complete compatible routes and keeps uncertified recipes dark.
+- [ ] Direct routes use gRPC streams; relayed routes carry the same versioned protobuf frames as opaque binary through the existing relay seam.
+- [ ] Existing request/work IDs, cancellation, Generation Telemetry, billing, and per-node attribution remain correlated.
+- [ ] No vLLM, Nakshatra, prima.cpp, or custom-engine control plane becomes a core dependency.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched
+- [ ] llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-009/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-003` must have `passes: true`; read `../evidence/DGR-003/README.md` and verify its referenced files/commands.
+- `DGR-008` must have `passes: true`; read `../evidence/DGR-008/README.md` and verify its referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

.scratch/distributed-gguf-runtime/issues/10-glm52-ornith-followup-audit.md

@@ -1,20 +0,0 @@
-# 10 — GLM-5.2 and Ornith follow-up support audit
-
-Status: pending
-
-## What to build
-
-Audit GLM-5.2 and Ornith after the smaller protocol smoke path and DeepSeek-V4-Flash audit. The output is a follow-up compatibility matrix focused on architecture/runtime blockers: DSA/MLA, hybrid attention, cache accounting, and GGUF/llama.cpp support.
-
-## Acceptance criteria
-
-- [ ] Verify GLM-5.2 PyTorch/Transformers serving requirements and cache semantics from primary model documentation.
-- [ ] Verify llama.cpp/GGUF support status for `glm_moe_dsa` or equivalent architecture support.
-- [ ] Verify Ornith/Qwen3.5-MoE and hybrid attention support status in the candidate runtimes.
-- [ ] Estimate artifact size and 128K cache memory by Shard range for each model.
-- [ ] Identify smallest quality-preserving quantization worth testing.
-- [ ] Convert each runtime blocker into a follow-up issue or upstream collaboration note.
-
-## Blocked by
-
-- 09 — DeepSeek-V4-Flash support audit.

.scratch/distributed-gguf-runtime/issues/10-pass-local-real-model-two-process-acceptance.md

@@ -0,0 +1,62 @@
+# 10 — Pass local real-model two-process acceptance
+
+Status: ready-for-agent
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-010` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a release engineer, I need real local distributed parity before involving network variability.
+
+## Expected durable outputs
+
+- Real local two-process commands and configuration
+- Raw parity, memory and performance results
+- evidence/DGR-010/README.md
+
+## Acceptance criteria
+
+- [ ] Two local worker processes open disjoint dense-Llama ranges from the certified Artifact.
+- [ ] Prefill and at least 32 greedy decode tokens match whole-model llama.cpp within the certified tolerance.
+- [ ] Each worker retains only its own tensors and Hot KV State.
+- [ ] Four concurrent Route Sessions pass isolation and cleanup checks.
+- [ ] Report TTFT, prefill/decode throughput, seam bytes/latency, worker RSS/VRAM, KV memory, batch size, and queue time.
+- [ ] Killing one worker produces a bounded structured failure rather than a deadlock.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched
+- [ ] llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched
+- [ ] Real-model execution is opt-in through MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 and records exact artifact/runtime/hardware evidence
+- [ ] Model artifacts remain on the configured mounted-drive storage and never under /home
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-010/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-009` must have `passes: true`; read `../evidence/DGR-009/README.md` and verify its referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

.scratch/distributed-gguf-runtime/issues/11-pass-a-real-heterogeneous-two-machine-route.md

@@ -0,0 +1,62 @@
+# 11 — Pass a real heterogeneous two-machine route
+
+Status: ready-for-agent
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-011` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a consumer-hardware operator, I need two physical machines to execute one GGUF model so that the distributed claim is real.
+
+## Expected durable outputs
+
+- Two-machine hardware/network/runtime manifest
+- Raw real-route metrics and output evidence
+- evidence/DGR-011/README.md
+
+## Acceptance criteria
+
+- [ ] Tracker selects two physical nodes with disjoint Shards and one exact certified recipe/compatibility class.
+- [ ] Actual CPU/GPU execution occurs on both nodes; synthetic workers do not satisfy acceptance.
+- [ ] Prefill/decode, concurrent-session isolation, telemetry, cancellation, and cleanup pass over the real transport/relay path.
+- [ ] Exact hardware, network, backend, model hash, route, commands, and raw metrics are recorded.
+- [ ] A model or recipe larger than one participating node's admitted memory is exercised when available.
+- [ ] Output drift is measured and incompatible mixed backends fail closed.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched
+- [ ] llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched
+- [ ] Real-model execution is opt-in through MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 and records exact artifact/runtime/hardware evidence
+- [ ] Model artifacts remain on the configured mounted-drive storage and never under /home
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-011/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-010` must have `passes: true`; read `../evidence/DGR-010/README.md` and verify its referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

.scratch/distributed-gguf-runtime/issues/12-implement-continuous-batching-and-bounded-admission.md

@@ -0,0 +1,63 @@
+# 12 — Implement continuous batching and bounded admission
+
+Status: ready-for-agent
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-012` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a node operator, I need active sessions batched safely so that concurrency increases aggregate throughput rather than serializing every request.
+
+## Expected durable outputs
+
+- Continuous batching/admission scheduler
+- Concurrency 1/2/4/8 report
+- Queue, batch and KV-pressure evidence
+- evidence/DGR-012/README.md
+
+## Acceptance criteria
+
+- [ ] Node scheduler admits sessions against weight, KV, scratch, and queue budgets.
+- [ ] Compatible decode steps from multiple sessions form llama.cpp batches while preserving per-session positions and outputs.
+- [ ] Prefill does not starve decode; scheduling policy and bounds are explicit.
+- [ ] Backpressure prevents unbounded queued activations or KV growth.
+- [ ] Capability telemetry reports active sessions, queue depth, batch occupancy, KV pressure, prefill/decode rates, and rejected admissions.
+- [ ] Concurrency 1/2/4/8 benchmark identifies saturation and shows no cross-session corruption.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched
+- [ ] llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-012/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-007` must have `passes: true`; read `../evidence/DGR-007/README.md` and verify its referenced files/commands.
+- `DGR-009` must have `passes: true`; read `../evidence/DGR-009/README.md` and verify its referenced files/commands.
+- `DGR-010` must have `passes: true`; read `../evidence/DGR-010/README.md` and verify its referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

.scratch/distributed-gguf-runtime/issues/13-harden-failure-cancellation-and-restart-semantics.md

@@ -0,0 +1,62 @@
+# 13 — Harden failure, cancellation, and restart semantics
+
+Status: ready-for-agent
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-013` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a client, I need failures to be bounded and explicit so that distributed speed does not come with hanging or corrupted generations.
+
+## Expected durable outputs
+
+- Failure/cancel/restart test matrix
+- Resource cleanup and billing-state evidence
+- evidence/DGR-013/README.md
+
+## Acceptance criteria
+
+- [ ] Deadlines and heartbeat/health loss terminate blocked stream operations.
+- [ ] Cancellation propagates across every Shard and releases local KV and queued buffers.
+- [ ] Duplicate steps are idempotent; uncertain mutations are never replayed silently.
+- [ ] Alpha failover restarts from token zero on a newly compatible route rather than importing unverified KV.
+- [ ] Worker death, stream reset, malformed bundle, stale epoch, and cache miss tests pass.
+- [ ] Billing/work records distinguish completed, cancelled, failed, and unverified work.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched
+- [ ] llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-013/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-008` must have `passes: true`; read `../evidence/DGR-008/README.md` and verify its referenced files/commands.
+- `DGR-009` must have `passes: true`; read `../evidence/DGR-009/README.md` and verify its referenced files/commands.
+- DGR-012 continuous batching is post-alpha; this story must harden single-session failure/cancellation first and later extend the same invariants to batched sessions.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

.scratch/distributed-gguf-runtime/issues/14-enforce-the-gguf-versus-safetensors-release-gate.md

@@ -0,0 +1,65 @@
+# 14 — Enforce the GGUF-versus-safetensors release gate
+
+Status: ready-for-agent
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-014` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As the product owner, I need an end-to-end comparison so that the native runtime ships only if it advances model access or performance.
+
+## Expected durable outputs
+
+- Immutable comparison against DGR-001 thresholds
+- Machine-readable final report
+- Ship/optimize/stop recommendation
+- evidence/DGR-014/README.md
+
+## Acceptance criteria
+
+- [ ] Run current distributed safetensors and distributed GGUF routes on the same certified model/hardware/network scenario where technically comparable.
+- [ ] Report quality, TTFT, prefill/decode throughput, aggregate concurrency throughput, p95 latency, seam cost, memory, KV pressure, failures, and cleanup.
+- [ ] Evaluate against the DGR-001 performance contract without changing thresholds after seeing results.
+- [ ] Ship recommendation is one of: promote GGUF, optimize a measured bottleneck with a new bounded task, or stop the native track.
+- [ ] Results clearly separate quantization gains from transport/runtime gains.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched
+- [ ] llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched
+- [ ] Real-model execution is opt-in through MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 and records exact artifact/runtime/hardware evidence
+- [ ] Model artifacts remain on the configured mounted-drive storage and never under /home
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-014/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-001` must have `passes: true`; read `../evidence/DGR-001/README.md` and verify its referenced files/commands.
+- `DGR-011` must have `passes: true`; read `../evidence/DGR-011/README.md` and verify its referenced files/commands.
+- `DGR-012` must have `passes: true`; read `../evidence/DGR-012/README.md` and verify its referenced files/commands.
+- `DGR-013` must have `passes: true`; read `../evidence/DGR-013/README.md` and verify its referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

.scratch/distributed-gguf-runtime/issues/15-add-and-certify-a-qwen3-qwen3-moe-adapter.md

@@ -0,0 +1,61 @@
+# 15 — Add and certify a Qwen3/Qwen3-MoE adapter
+
+Status: ready-for-agent
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-015` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a client seeking top models, I need a separately certified MoE-capable architecture after the dense runtime proves stable.
+
+## Expected durable outputs
+
+- Qwen3-family architecture adapter
+- Architecture-specific parity/admission/performance results
+- evidence/DGR-015/README.md
+
+## Acceptance criteria
+
+- [ ] Implement explicit tensor ownership, router/top-k, expert/shared-expert, Q/K normalization, boundary bundle, and cache semantics for the selected Qwen3 family recipe.
+- [ ] Do not reuse the dense-Llama adapter through unchecked name substitutions.
+- [ ] Whole-model versus distributed prefill/decode parity passes the architecture-specific tolerance.
+- [ ] Expert memory ownership and communication are measured.
+- [ ] Real consumer-hardware acceptance and capability admission pass before the recipe becomes routable.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched
+- [ ] llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched
+- [ ] Real-model execution is opt-in through MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 and records exact artifact/runtime/hardware evidence
+- [ ] Model artifacts remain on the configured mounted-drive storage and never under /home
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-015/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-014` must have `passes: true`; read `../evidence/DGR-014/README.md` and verify its referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

.scratch/distributed-gguf-runtime/issues/16-produce-the-upstream-llama-cpp-collaboration-package.md

@@ -0,0 +1,60 @@
+# 16 — Produce the upstream llama.cpp collaboration package
+
+Status: ready-for-agent
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-016` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a maintainer, I need narrow upstreamable proposals so that our patch burden can shrink without asking llama.cpp to own Meshnet networking.
+
+## Expected durable outputs
+
+- Narrow upstream patches/tests
+- Generic API design note
+- Human-ready llama.cpp outreach package
+- evidence/DGR-016/README.md
+
+## Acceptance criteria
+
+- [ ] Separate generic llama.cpp hooks from Meshnet protocol/control-plane code.
+- [ ] Prepare minimal reproducible examples and tests for range-aware loading, boundary input/output, and layer-filtered KV.
+- [ ] Compare the proposal with Nakshatra and prima.cpp evidence and explain why the API is generally useful.
+- [ ] Preserve one scoped commit/patch per concern against the exact upstream pin.
+- [ ] Produce an outreach document suitable for Georgi/llama.cpp maintainers; actual sending remains a human action.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched
+- [ ] llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-016/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-010` must have `passes: true`; read `../evidence/DGR-010/README.md` and verify its referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

.scratch/distributed-gguf-runtime/issues/18-certify-whole-model-glm-5-2-runtime-semantics.md

@@ -0,0 +1,67 @@
+# 18 — Certify whole-model GLM-5.2 runtime semantics
+
+Status: blocked (2026-07-14) — no 256-GiB-class host available
+
+> **Blocked:** This story requires a 256-GiB-class host with at least 224 GiB
+> runtime-accessible memory and 250 GB free storage outside `/home`. The
+> development host has 124.9 GiB MemTotal and no eligible filesystem (largest:
+> 74.2 GB free). Exact preflight output is preserved in
+> [evidence/DGR-018/BLOCKED.md](../evidence/DGR-018/BLOCKED.md); preflight
+> scripts were preserved at commit a0f28b5 (`scripts/glm_whole_model_preflight.py`,
+> `scripts/verify_glm_shards.py`). Resume by re-running the preflight on a
+> qualifying host — do not substitute a smaller model.
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md), [GLM-5.2-MAX-ALPHA-ROADMAP.md](../GLM-5.2-MAX-ALPHA-ROADMAP.md), and this issue completely before changing code.
+- This issue is `DGR-018` in [prd.json](../prd.json).
+- Read and verify every dependency evidence README.
+- Inspect current upstream behavior and `git status`; community claims that a model “works” are not correctness evidence.
+
+## Description
+
+As a runtime maintainer, I need a certified whole-model oracle for the exact lowest-quant GLM-5.2 artifact so that distributed parity is measured against correct MoE, DSA, IndexShare, cache, and Max-template semantics.
+
+## Expected durable outputs
+
+- Verified target artifact on mounted storage
+- Stock-pin baseline and warning/tensor inventory
+- Focused GLM runtime correctness tests and minimum patch series, if required
+- Signed whole-model oracle recipe, deterministic outputs, metrics, and limitations
+- `evidence/DGR-018/README.md`
+
+## Acceptance criteria
+
+- [ ] Preflight a 256-GiB-class reference host with at least 224 GiB runtime-accessible memory after OS reservation and approximately 250 GB free mounted storage; abort before download if storage resolves under `/home` or requirements are unmet.
+- [ ] Configure the oracle's alpha lane for 16,384 context, concurrency 1, and Q8_0 MLA/indexer KV; record actual MLA, indexer-cache, scratch, and peak resident allocations separately.
+- [ ] Download/resume and verify all six DGR-017 `UD-IQ1_S` files against exact sizes and LFS SHA-256 values.
+- [ ] Load the target with the unmodified DGR-004 pin first and retain full logs, tensor warnings, peak memory, context/KV allocation, TTFT, and decode timing.
+- [ ] Prove from graph/runtime evidence whether 256-expert MoE/top-8/shared expert, DSA lightning indexer/sparse attention, IndexShare Full/Shared reuse, and `reasoning_effort=max` are active.
+- [ ] Dense-attention or replicated-indexer compatibility fallback is labeled incomplete and cannot become the oracle solely because it emits plausible text.
+- [ ] Audit focused upstream/Mesh-LLM donor tests and patches; adopt or rewrite only independently understood minimum changes and record provenance/rejection rationale.
+- [ ] Handle the trailing NextN/MTP tensors explicitly; MTP may remain disabled for alpha, but ignored tensors and layer-count behavior must be understood and tested.
+- [ ] Run deterministic prefill/greedy decode and fixed Max-mode coding, structured-output/tool-call, and reasoning sentinels; retain raw prompts, outputs, token IDs, and compared state/logit evidence.
+- [ ] Bind the oracle to exact artifact, tokenizer/template, adapter, backend class, compute/activation/KV dtypes, context/RoPE, llama.cpp commit, and patch-series hash.
+- [ ] Produce a signed `pass` or `stop` semantic-runtime verdict; no performance threshold is weakened after execution.
+- [ ] Targeted native/unit tests and clean pinned rebuild pass.
+- [ ] `git diff --check` passes.
+- [ ] Default tests remain model-download-free/GPU-free; full-target execution is opt-in and never runs in default CI.
+- [ ] Model artifacts and caches remain on configured mounted-drive storage and never under `/home`.
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story.
+- [ ] Write `evidence/DGR-018/README.md` with exact hardware, storage, source revisions, files, commands, raw-result paths, real results, limitations, and handoff.
+- [ ] Update only this story issue to `Status: done` after every acceptance criterion passes.
+
+## Dependency handoff
+
+- `DGR-003`, `DGR-004`, and `DGR-017` must have `passes: true`; read and verify their evidence READMEs.
+
+## Finish contract
+
+- If a 256-GiB-class host with at least 224 GiB runtime-accessible memory is unavailable, write `evidence/DGR-018/BLOCKED.md` with exact preflight output; do not substitute a smaller model.
+- Preserve real failures and blockers; never fabricate output.
+- Emit `<promise>COMPLETE</promise>` only after the signed oracle evidence exists.
+
+## References
+
+- [GLM-5.2 Max alpha roadmap](../GLM-5.2-MAX-ALPHA-ROADMAP.md)
+- [llama.cpp GLM-5.2 issue](https://github.com/ggml-org/llama.cpp/issues/24730)

.scratch/distributed-gguf-runtime/issues/19-implement-and-certify-glm-5-2-range-dsa-indexshare.md

@@ -0,0 +1,63 @@
+# 19 — Implement and certify GLM-5.2 range, DSA, and IndexShare semantics
+
+Status: blocked (2026-07-14) — waiting on DGR-018
+
+> **Blocked:** Depends on DGR-018's whole-model IQ1_S oracle, which is blocked
+> on a 256-GiB-class host (≥ 224 GiB runtime-accessible memory). See
+> [evidence/DGR-018/BLOCKED.md](../evidence/DGR-018/BLOCKED.md). Locked
+> fixture/target parity cannot be certified without that oracle.
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md), [GLM-5.2-MAX-ALPHA-ROADMAP.md](../GLM-5.2-MAX-ALPHA-ROADMAP.md), and this issue completely before changing code.
+- This issue is `DGR-019` in [prd.json](../prd.json).
+- Read and verify every dependency evidence README.
+- Inspect current source and `git status`; do not generalize through tensor-name substitution.
+
+## Description
+
+As a target-model operator, I need explicit range-owned GLM-5.2 semantics so that contiguous consumer Shards preserve the whole-model MoE, DSA, IndexShare, and local-KV computation.
+
+## Expected durable outputs
+
+- Explicit GLM-5.2 architecture adapter and tensor-ownership rules
+- IndexShare-aware Shard planner and named DSA sideband protocol mapping
+- GLM architecture fixture and target parity evidence
+- Memory/tensor/KV ownership report
+- `evidence/DGR-019/README.md`
+
+## Acceptance criteria
+
+- [ ] Implement explicit ownership for 78 main layers, head embedding, tail norm/output head, routed/shared experts, DSA/indexer tensors, and the certified NextN/MTP policy.
+- [ ] Each contiguous layer owner keeps all 256 experts and the shared-expert path for its layers local; no public cross-machine expert collectives are introduced.
+- [ ] Define and fixture-test compressed MLA KV ownership for locally owned layers, keyed by Route Session/epoch, for DGR-007 to implement.
+- [ ] Implement native DSA lightning indexer/top-2,048 and sparse-attention graph behavior matching DGR-018.
+- [ ] Parse and validate artifact `indexer_types`; implement Full producer and Shared consumer behavior without fabricated/duplicated indexer tensors.
+- [ ] Prefer Shard boundaries that preserve IndexShare ownership groups; when memory fit forces a split, carry a typed, bounded, validated top-k sideband in the DGR-006 named bundle.
+- [ ] Reject missing, stale, wrong-width, wrong-position, or Shared-before-Full sideband/index state.
+- [ ] Demonstrate mapped/resident weights and allocated KV scale with exact owned tensors/layers rather than the full model.
+- [ ] Same-host two-stage F32 seam fixture produces 32 exact greedy tokens against the DGR-018 semantics; production seam meets the pre-locked token/similarity thresholds.
+- [ ] If full-target same-host execution cannot fit, use a layer-reduced GLM architecture fixture only for graph parity and defer full-artifact parity to DGR-020; label it non-target evidence.
+- [ ] Add deterministic fixture tests for MoE route/shared expert, DSA Full/Shared groups, internal group split, endpoint ownership, KV filtering, malformed metadata, and NextN policy.
+- [ ] Targeted pytest/CTest/native tests pass; pinned patch stack applies and rebuilds cleanly.
+- [ ] `python -m compileall packages tests` and `git diff --check` pass.
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free.
+- [ ] Full deterministic `pytest -q` passes, or exact pre-existing unrelated failures are recorded.
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story.
+- [ ] Write `evidence/DGR-019/README.md` with files, commands, real results, raw evidence, limitations, and dependent-story handoff.
+- [ ] Update only this story issue to `Status: done` after every acceptance criterion passes.
+
+## Dependency handoff
+
+- `DGR-005`, `DGR-006`, and `DGR-018` must have `passes: true`; read and verify their evidence READMEs.
+- DGR-007 integration may be completed in parallel only after the adapter's cache contract is explicit.
+
+## Finish contract
+
+- Preserve real failures and blockers; never claim target parity from the architecture fixture.
+- Emit `<promise>COMPLETE</promise>` only after evidence exists.
+
+## References
+
+- [GLM-5.2 Max alpha roadmap](../GLM-5.2-MAX-ALPHA-ROADMAP.md)
+- [Current architecture](../architecture.md)

.scratch/distributed-gguf-runtime/issues/20-pass-real-distributed-glm-5-2-max-alpha.md

@@ -0,0 +1,67 @@
+# 20 — Pass real distributed GLM-5.2 Max alpha acceptance
+
+Status: blocked (2026-07-14) — waiting on DGR-018/DGR-019
+
+> **Blocked:** Depends on DGR-018 and DGR-019, both blocked on the 256-GiB-class
+> oracle host (≥ 224 GiB runtime-accessible memory), plus enough physical
+> consumer nodes that no single node admits the whole recipe. See
+> [evidence/DGR-018/BLOCKED.md](../evidence/DGR-018/BLOCKED.md).
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md), [GLM-5.2-MAX-ALPHA-ROADMAP.md](../GLM-5.2-MAX-ALPHA-ROADMAP.md), and this issue completely before changing code.
+- This issue is `DGR-020` in [prd.json](../prd.json).
+- Read and verify every dependency evidence README and the immutable DGR-017 contract.
+- Inspect live source, hardware, network, storage, and `git status`; synthetic or layer-reduced evidence cannot satisfy this story.
+
+## Description
+
+As an alpha user, I need the exact lowest-quant GLM-5.2 model to run in Max reasoning mode across consumer machines through Meshnet so that the project has proven its end goal on real hardware.
+
+## Expected durable outputs
+
+- Exact multi-node hardware/network/storage/runtime manifest
+- Signed target identity, ownership, parity, quality, performance, and reliability reports
+- Raw worker/tracker/API logs, metrics, prompts, outputs, and cleanup evidence
+- Explicit immutable `alpha` or `stop` verdict
+- `evidence/DGR-020/README.md`
+
+## Acceptance criteria
+
+- [ ] Use the exact DGR-017 `UD-IQ1_S` artifact and certified DGR-018/DGR-019 runtime recipe with all source and patch hashes verified.
+- [ ] Use at least two physical consumer machines; each reserves at least `max(20% of physically usable memory, 8 GiB)` outside weight-plus-Q8-KV placement, no node can place the complete recipe, unified memory is counted once, and measured peak use stays within physical memory without swap/overcommit.
+- [ ] Treat 5×64 GiB or 3×96/128 GiB as the recommended topology; arithmetic-minimum 4×64 or 2×128 qualifies only with exact contiguous placement and measured reserve evidence.
+- [ ] Use a same-switch wired network with at least 2.5 GbE; record 10 GbE as recommended and measure hop RTT/serialization/queue latency.
+- [ ] Tracker selects disjoint contiguous Shards whose exact required tensor inventory has complete union and no unintended overlap.
+- [ ] Every stage reports real CPU/GPU compute, owned tensor bytes/layers, local KV, backend, queue, and seam telemetry; synthetic, passthrough, or zero-layer workers fail acceptance.
+- [ ] Runtime evidence proves native GLM MoE/shared expert, DSA lightning indexer/sparse attention, and IndexShare Full/Shared paths are active; dense fallback fails acceptance.
+- [ ] OpenAI-compatible API applies and records `reasoning_effort=max`, stable model ID, finish reason, and token usage.
+- [ ] Configure 16,384 context, concurrency 1, and Q8_0 MLA/indexer KV; complete the fixed 4,096-token prefill lane and generate at least 512 output tokens or valid natural EOS after at least 128 tokens.
+- [ ] Production seam meets at least 0.90 greedy token agreement and 0.999 mean compared-state/logit cosine similarity against DGR-018 on the fixed corpus, with no non-finite or malformed tensors.
+- [ ] Fixed coding, structured tool-call/JSON, and multi-step reasoning sentinels produce parseable relevant outputs retained for human review.
+- [ ] After warm-up, fixed Max lane median decode is at least 0.5 token/s, 4,096-token-prompt TTFT is at most 10 minutes, and no unexplained stall exceeds 60 seconds without progress telemetry.
+- [ ] Record per-stage/seam p50/p95 latency, bytes, compute time, peak RSS/VRAM, KV pressure, network properties, errors, and total energy when available.
+- [ ] Two consecutive cold starts load, generate, release, and exit without leaked processes, mapped weights, queues, or KV leases.
+- [ ] Cancellation in prefill and decode releases all stages; one worker loss aborts the route; retry starts from token zero on a newly compatible route; stale epochs and duplicate steps fail closed.
+- [ ] Model artifacts/caches remain on configured mounted-drive storage and never under `/home`; secret scan passes.
+- [ ] Run targeted/full deterministic tests and clean native rebuild in addition to target execution; `git diff --check` passes.
+- [ ] Preserve all raw failures and do not weaken DGR-017 thresholds after results.
+- [ ] Write a signed `alpha` verdict only if every criterion passes; otherwise write `stop` with the measured bottleneck and next narrow action.
+- [ ] Write `evidence/DGR-020/README.md` with exact commands, manifests, raw-result paths, outputs, limitations, and post-alpha handoff.
+- [ ] Update only this story issue to `Status: done` after the immutable verdict and all evidence exist.
+
+## Dependency handoff
+
+- `DGR-007`, `DGR-008`, `DGR-009`, `DGR-011`, `DGR-013`, `DGR-017`, `DGR-018`, and `DGR-019` must have `passes: true`; read and verify their evidence READMEs.
+- DGR-012 continuous batching and DGR-014 final comparative release gate are post-alpha and do not block one target session.
+
+## Finish contract
+
+- If required physical nodes/storage are unavailable, write `evidence/DGR-020/BLOCKED.md`; do not substitute a smaller model, API, synthetic worker, or single host.
+- Preserve real failures and blockers; never fabricate target output or telemetry.
+- Emit `<promise>COMPLETE</promise>` only after the signed verdict exists.
+
+## References
+
+- [GLM-5.2 Max alpha roadmap](../GLM-5.2-MAX-ALPHA-ROADMAP.md)
+- [Ralph execution context](../RALPH-CONTEXT.md)

.scratch/distributed-gguf-runtime/milestones.md

@@ -1,32 +1,47 @@
-# Distributed GGUF Runtime Milestones
+# Distributed GGUF runtime milestones
 
-## Proposed Breakdown
+The exact alpha target and immutable acceptance gates are defined in [GLM-5.2-MAX-ALPHA-ROADMAP.md](GLM-5.2-MAX-ALPHA-ROADMAP.md).
 
-| Order | Issue | Title | Blocked by | User-visible proof |
-|---:|---|---|---|---|
-| 1 | [01](./issues/01-route-session-lifecycle.md) | Route Session lifecycle | None | Stable route/session status and cleanup |
-| 2 | [02](./issues/02-prefill-decode-binary-http.md) | Prefill/decode binary HTTP protocol | 01 | Stub route proves prefill chunks and one-step decode payloads |
-| 3 | [03](./issues/03-generation-telemetry-and-streaming.md) | Generation Telemetry and streaming response contract | 01 | Client sees route progress and streamed deltas when available |
-| 4 | [04](./issues/04-pytorch-distributed-kv-reference.md) | PyTorch distributed KV reference route | 01, 02, 03 | Distributed PyTorch decode stops full-prompt recompute |
-| 5 | [05](./issues/05-local-llamacpp-gguf-backend.md) | Local llama.cpp/GGUF backend | None | Local GGUF model serves through node API |
-| 6 | [06](./issues/06-model-artifact-manifest.md) | Model Artifact manifest and Shard advertisement | 01 | Node verifies artifacts and advertises serveable Shards |
-| 7 | [07](./issues/07-llamacpp-layer-boundary-prototype.md) | llama.cpp layer-boundary prototype | 02, 05, 06 | Local two-process GGUF route identifies upstream API |
-| 8 | [08](./issues/08-networked-distributed-gguf-route.md) | Networked distributed GGUF route | 03, 04, 06, 07 | Two machines serve one GGUF route with telemetry |
-| 9 | [09](./issues/09-deepseek-v4-flash-support-audit.md) | DeepSeek-V4-Flash support audit | None | Runtime recommendation for first serious large model |
-| 10 | [10](./issues/10-glm52-ornith-followup-audit.md) | GLM-5.2 and Ornith follow-up support audit | 09 | Follow-up compatibility matrix and upstream blockers |
+## Completed foundation
 
-## First Three To Implement
+- DGR-001 locks the safetensors-versus-GGUF performance/fit/quality contract. Its immutable CPU v1 verdict remains `stop`; separate ROCm evidence supports a fit-oriented investigation without rewriting CPU evidence.
+- DGR-002 defines the backend-neutral gRPC/Protobuf Shard protocol.
 
-1. **01 — Route Session lifecycle**: makes every later cache, telemetry, and route decision concrete.
-2. **02 — Prefill/decode binary HTTP protocol**: proves the payload shape and route/session headers before model internals.
-3. **03 — Generation Telemetry and streaming response contract**: gives every later long-running route a visible user experience and failure surface.
+## Gate A — exact GLM-5.2 target and oracle
 
-## Parallel Work
+- DGR-017 locks official/GGUF revisions, `UD-IQ1_S`, hashes, resource accounting, Max-mode semantics, and alpha thresholds.
+- DGR-003 builds exact recipe identity on DGR-002 and DGR-017.
+- DGR-004 creates the reproducible pinned llama.cpp boundary after stock GLM behavior is measured.
+- DGR-018 certifies one whole-model `UD-IQ1_S` oracle with real MoE, DSA, IndexShare, KV, and `reasoning_effort=max` semantics. **Blocked (2026-07-14):** requires a 256-GiB-class host (≥ 224 GiB runtime-accessible memory); see [evidence/DGR-018/BLOCKED.md](evidence/DGR-018/BLOCKED.md). DGR-019 and DGR-020 are blocked transitively.
 
-- **05 — Local llama.cpp/GGUF backend** can run in parallel with 01–03 because it is a full-model local backend.
-- **09 — DeepSeek-V4-Flash support audit** can run in parallel because it is research/compatibility work.
+## Gate B — minimal native execution seam
 
-## Human-Gated Work
+- DGR-005 implements range-owned tensors using dense Llama as a cheap structural fixture.
+- DGR-006 proves the generic named activation boundary and F32 correctness lane.
+- DGR-019 implements explicit GLM-5.2 MoE/MLA/DSA/IndexShare/NextN range semantics and parity.
 
-- **07 — llama.cpp layer-boundary prototype** is the collaboration point with Georgi/upstream llama.cpp.
-- **08 — Networked distributed GGUF route** should wait until the PyTorch reference route proves the cache/session contract.
+## Gate C — native Meshnet route
+
+- DGR-007 isolates Shard-local Hot KV State.
+- DGR-008 exposes the native worker over gRPC.
+- DGR-009 integrates the worker without replacing Meshnet's control plane.
+- DGR-010 passes local real-model two-process acceptance.
+- DGR-011 passes real two-physical-machine execution.
+- DGR-013 supplies the cancellation/node-loss/restart/cleanup subset required by alpha.
+
+## Gate D — GLM-5.2 Max alpha verdict
+
+- DGR-020 runs the exact `UD-IQ1_S` target across enough physical consumer nodes that no one node can admit the whole recipe.
+- It produces an immutable `alpha` or `stop` verdict from target identity, native GLM semantics, parity, Max-mode usefulness, minimum speed, telemetry, failure, and cleanup evidence.
+- Synthetic workers, layer-reduced fixtures, dense-attention fallbacks, and single-host execution cannot satisfy this gate.
+
+## Gate E — post-alpha product hardening
+
+- DGR-012 adds continuous batching and bounded admission.
+- DGR-014 compares distributed GGUF against the current reference route under locked thresholds.
+- Longer contexts progress through 32K/128K/200K before 1M certification.
+- MTP/speculative decoding and source-bound layer packages remain measured optimizations.
+- DGR-016 prepares narrow upstream llama.cpp collaboration material.
+- DGR-015 adds Qwen3/Qwen3-MoE only as later architecture expansion.
+
+No later gate may be claimed from synthetic workers or documentation-only evidence. Model artifacts remain on mounted-drive storage and never under `/home`.

.scratch/distributed-gguf-runtime/prd.json

@@ -0,0 +1,655 @@
+{
+  "name": "Performant Concurrent Distributed GGUF Runtime",
+  "branchName": "ralph/performant-concurrent-distributed-gguf",
+  "description": "Build a performant concurrent distributed GGUF runtime on Meshnet. Alpha is gated by exact GLM-5.2 UD-IQ1_S execution in reasoning_effort=max mode across consumer machines; synthetic or dense-fallback evidence cannot satisfy the target.",
+  "userStories": [
+    {
+      "id": "DGR-001",
+      "title": "Lock the safetensors-versus-GGUF performance contract",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/01-lock-the-safetensors-versus-gguf-performance-contract.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a runtime engineer, I need a controlled baseline so that GGUF work proceeds from measured speed, memory, and quality rather than reputation.",
+      "acceptanceCriteria": [
+        "Benchmark the same model architecture/revision, machine, prompts, context lengths, output lengths, sampling policy, and concurrency across the current Transformers/safetensors recipe and whole-model llama.cpp recipes.",
+        "Separate correctness/quality lanes from quantized performance/fit lanes instead of claiming BF16 and Q4 are numerically equivalent.",
+        "Report TTFT, prefill tok/s, decode tok/s, p50/p95 latency, aggregate throughput, RSS, VRAM, artifact size, failures, and output drift in machine-readable JSON.",
+        "Add concurrency levels 1 and 4 where memory permits.",
+        "Write a versioned performance contract consumed by later release gates, including an explicit stop condition when llama.cpp/GGUF has no meaningful speed or fit benefit.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Real-model execution is opt-in through MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 and records exact artifact/runtime/hardware evidence",
+        "Model artifacts remain on the configured mounted-drive storage and never under /home",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-001/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 1,
+      "passes": true,
+      "notes": "Source issue: docs/issues/distributed-gguf-runtime/01-lock-the-safetensors-versus-gguf-performance-contract.md (moved on close, MAINT-003)",
+      "dependsOn": []
+    },
+    {
+      "id": "DGR-002",
+      "title": "Adopt the versioned gRPC Shard protocol",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/02-adopt-the-versioned-grpc-shard-protocol.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a node developer, I need a battle-proven streaming protocol so that Python and C++ Shards communicate without a custom socket protocol.",
+      "acceptanceCriteria": [
+        "Add a Protocol Buffers schema for capability, health, session stream, release, and cancellation operations.",
+        "Define one long-lived bidirectional gRPC stream per Route Session Activation Seam with deadlines, cancellation, flow control, and structured errors.",
+        "Define bounded chunking for prefill and a small decode fast path.",
+        "Carry schema version, request/work ID, Route Session ID, route epoch, artifact/recipe fingerprint, Shard range/effective start, phase, position, idempotency step, cache expectation, compression, and checksum.",
+        "Define a versioned named-tensor bundle with per-tensor name, shape, dtype, byte order, and payload fragments.",
+        "Add generated-schema round-trip and compatibility tests in Python and C++.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-002/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 2,
+      "passes": true,
+      "notes": "Source issue: docs/issues/distributed-gguf-runtime/02-adopt-the-versioned-grpc-shard-protocol.md (moved on close, MAINT-003)",
+      "dependsOn": []
+    },
+    {
+      "id": "DGR-003",
+      "title": "Define exact Artifact and runtime recipe identity",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/03-define-exact-artifact-and-runtime-recipe-identity.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs the Tracker, I need exact compatibility identity so that only numerically and operationally compatible Shards form an Inference Route.",
+      "acceptanceCriteria": [
+        "Separate weight quantization, activation dtype, compute dtype, KV dtype/layout, tokenizer revision, architecture adapter, backend, and runtime version.",
+        "Bind derivative or split artifacts to an exact source Model Artifact hash and Shard range.",
+        "Produce a stable compatibility fingerprint used by capability admission and the gRPC handshake.",
+        "Fail closed on mismatched artifact, tokenizer, architecture, range, boundary schema, activation recipe, or cache layout.",
+        "Keep unsupported recipes registered-but-dark until a real distributed forward certifies them.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-003/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 4,
+      "passes": true,
+      "notes": "DGR-003-emission: native loaded-artifact adapter derives exact identity from immutable GGUF/runtime inputs, doctor emits it only for that adapter, and SessionOpen fails closed before acceptance. Exact identities remain tracker-uncertified and dark until a real distributed forward is certified. (moved on close, MAINT-003)",
+      "dependsOn": [
+        "DGR-002",
+        "DGR-017"
+      ]
+    },
+    {
+      "id": "DGR-004",
+      "title": "Create the reproducible pinned llama.cpp patch stack",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/04-create-the-reproducible-pinned-llama-cpp-patch-stack.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a maintainer, I need a small auditable fork boundary so that upstream updates do not turn the runtime into an unmaintainable stitched codebase.",
+      "acceptanceCriteria": [
+        "Pin one exact llama.cpp commit through a reproducible source dependency mechanism.",
+        "Store a numbered minimal patch stack separately from Meshnet networking code.",
+        "Add a build script that applies/checks patches and builds the standalone worker without manual source copying.",
+        "Record upstream file/ABI assumptions and fail clearly when the pin changes.",
+        "Preserve upstream license and attribution notices.",
+        "Add a clean rebuild smoke test that does not download a model.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-004/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 5,
+      "passes": true,
+      "notes": "Source issue: docs/issues/distributed-gguf-runtime/04-create-the-reproducible-pinned-llama-cpp-patch-stack.md (moved on close, MAINT-003)",
+      "dependsOn": [
+        "DGR-001",
+        "DGR-017"
+      ]
+    },
+    {
+      "id": "DGR-005",
+      "title": "Implement dense-Llama range-aware GGUF ownership",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/05-implement-dense-llama-range-aware-gguf-ownership.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a node, I need to map only my assigned dense-Llama Shard so that aggregate consumer memory can hold a model larger than one node.",
+      "acceptanceCriteria": [
+        "Register and allocate only `blk.N.*` tensors in the assigned range.",
+        "Load embeddings only for the head and final norm/LM head only for the tail, including tied embeddings.",
+        "Prefer range-aware mapping from one exact source GGUF; if derivative sub-GGUFs are used temporarily, verify source/slice hashes and avoid claiming final artifact semantics.",
+        "Report authoritative loaded range and endpoint ownership from the model, not operator CLI claims.",
+        "Demonstrate mapped/resident memory scales with owned tensors rather than full model size.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-005/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 6,
+      "passes": true,
+      "notes": "Source issue: docs/issues/distributed-gguf-runtime/05-implement-dense-llama-range-aware-gguf-ownership.md (moved on close, MAINT-003)",
+      "dependsOn": [
+        "DGR-003",
+        "DGR-004"
+      ]
+    },
+    {
+      "id": "DGR-006",
+      "title": "Implement architecture-defined boundary input/output",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/06-implement-architecture-defined-boundary-input-output.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a Shard, I need to consume and emit the correct transformer boundary state so that disjoint processes reproduce whole-model execution.",
+      "acceptanceCriteria": [
+        "Head accepts token IDs and owns token embedding.",
+        "Middle/tail bypass token embedding and accept the named boundary bundle.",
+        "Amend the decode fast path so `DecodeStep` carries a versioned `TensorBundle`, not one `NamedTensor`; retain a compact one-tensor encoding and fallback compatibility, regenerate Python/C++ schema code, and update protocol goldens.",
+        "Define an explicit typed tail result for logits and/or sampled token output, with sampling parameters and chat-template/reasoning mode bound to exact request/recipe identity.",
+        "Non-tail emits the unnormalized architecture-defined residual/boundary before final norm/head and before tail-only row pruning.",
+        "Tail emits logits or token output through an explicit sampling contract.",
+        "Dense-Llama whole-model versus two-range prefill and greedy-decode parity passes the documented tolerance.",
+        "The adapter interface fails closed for uncertified architectures.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-006/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 7,
+      "passes": true,
+      "notes": "Source issue: docs/issues/distributed-gguf-runtime/06-implement-architecture-defined-boundary-input-output.md (moved on close, MAINT-003)",
+      "dependsOn": [
+        "DGR-002",
+        "DGR-005"
+      ]
+    },
+    {
+      "id": "DGR-007",
+      "title": "Add isolated concurrent local Hot KV State",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/07-add-isolated-concurrent-local-hot-kv-state.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a client, I need concurrent Route Sessions to retain independent per-Shard cache so that one request cannot clear or corrupt another.",
+      "acceptanceCriteria": [
+        "Map `(Route Session ID, route epoch)` to an isolated llama sequence or bounded context.",
+        "Allocate KV only for owned layers.",
+        "Support prefill append, decode append, truncate, release, TTL/LRU eviction, and explicit cache-miss response.",
+        "Reject stale epochs and incompatible cache recipes.",
+        "At least four concurrent sessions on a small model complete without token or KV cross-talk.",
+        "Cancellation/release of one session leaves other sessions intact and memory returns to the configured budget.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-007/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 10,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/07-add-isolated-concurrent-local-hot-kv-state.md",
+      "dependsOn": [
+        "DGR-006",
+        "DGR-019"
+      ]
+    },
+    {
+      "id": "DGR-008",
+      "title": "Build the standalone C++ gRPC Shard worker",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/08-build-the-standalone-c-grpc-shard-worker.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a node runtime, I need one supervised native process so that llama.cpp internals remain behind a stable project-owned protocol.",
+      "acceptanceCriteria": [
+        "Worker exposes capability, health, session stream, release, cancellation, and metrics services from DGR-002.",
+        "Worker loads one exact Artifact/recipe/Shard identity and refuses mismatched requests.",
+        "Streaming path enforces bounded messages, flow control, deadlines, idempotency, and independent session cancellation.",
+        "Worker does not expose raw llama.cpp RPC or arbitrary GGML graph execution.",
+        "Graceful shutdown releases sessions; crash behavior is bounded and observable.",
+        "Python integration tests run against a fake model mode without model downloads.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-008/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 11,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/08-build-the-standalone-c-grpc-shard-worker.md",
+      "dependsOn": [
+        "DGR-002",
+        "DGR-003",
+        "DGR-004",
+        "DGR-006",
+        "DGR-007"
+      ]
+    },
+    {
+      "id": "DGR-009",
+      "title": "Integrate the native worker with Meshnet",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/09-integrate-the-native-worker-with-meshnet.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs the existing node service, I need a GGUF Shard backend adapter so that the Tracker, relay, billing, telemetry, and capability admission remain the sole control plane.",
+      "acceptanceCriteria": [
+        "Implement the existing model-backend surface without changing Transformers behavior.",
+        "Registration carries exact validated GGUF recipe, Shard, backend and concurrency/KV capacity.",
+        "Tracker forms only complete compatible routes and keeps uncertified recipes dark.",
+        "Direct routes use gRPC streams; relayed routes carry the same versioned protobuf frames as opaque binary through the existing relay seam.",
+        "Existing request/work IDs, cancellation, Generation Telemetry, billing, and per-node attribution remain correlated.",
+        "No vLLM, Nakshatra, prima.cpp, or custom-engine control plane becomes a core dependency.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-009/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 12,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/09-integrate-the-native-worker-with-meshnet.md",
+      "dependsOn": [
+        "DGR-003",
+        "DGR-008"
+      ]
+    },
+    {
+      "id": "DGR-010",
+      "title": "Pass local real-model two-process acceptance",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/10-pass-local-real-model-two-process-acceptance.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a release engineer, I need real local distributed parity before involving network variability.",
+      "acceptanceCriteria": [
+        "Two local worker processes open disjoint dense-Llama ranges from the certified Artifact.",
+        "Prefill and at least 32 greedy decode tokens match whole-model llama.cpp within the certified tolerance.",
+        "Each worker retains only its own tensors and Hot KV State.",
+        "Four concurrent Route Sessions pass isolation and cleanup checks.",
+        "Report TTFT, prefill/decode throughput, seam bytes/latency, worker RSS/VRAM, KV memory, batch size, and queue time.",
+        "Killing one worker produces a bounded structured failure rather than a deadlock.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Real-model execution is opt-in through MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 and records exact artifact/runtime/hardware evidence",
+        "Model artifacts remain on the configured mounted-drive storage and never under /home",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-010/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 13,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/10-pass-local-real-model-two-process-acceptance.md",
+      "dependsOn": [
+        "DGR-009"
+      ]
+    },
+    {
+      "id": "DGR-011",
+      "title": "Pass a real heterogeneous two-machine route",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/11-pass-a-real-heterogeneous-two-machine-route.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a consumer-hardware operator, I need two physical machines to execute one GGUF model so that the distributed claim is real.",
+      "acceptanceCriteria": [
+        "Tracker selects two physical nodes with disjoint Shards and one exact certified recipe/compatibility class.",
+        "Actual CPU/GPU execution occurs on both nodes; synthetic workers do not satisfy acceptance.",
+        "Prefill/decode, concurrent-session isolation, telemetry, cancellation, and cleanup pass over the real transport/relay path.",
+        "Exact hardware, network, backend, model hash, route, commands, and raw metrics are recorded.",
+        "A model or recipe larger than one participating node's admitted memory is exercised when available.",
+        "Output drift is measured and incompatible mixed backends fail closed.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Real-model execution is opt-in through MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 and records exact artifact/runtime/hardware evidence",
+        "Model artifacts remain on the configured mounted-drive storage and never under /home",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-011/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 14,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/11-pass-a-real-heterogeneous-two-machine-route.md",
+      "dependsOn": [
+        "DGR-010"
+      ]
+    },
+    {
+      "id": "DGR-012",
+      "title": "Implement continuous batching and bounded admission",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/12-implement-continuous-batching-and-bounded-admission.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a node operator, I need active sessions batched safely so that concurrency increases aggregate throughput rather than serializing every request.",
+      "acceptanceCriteria": [
+        "Node scheduler admits sessions against weight, KV, scratch, and queue budgets.",
+        "Compatible decode steps from multiple sessions form llama.cpp batches while preserving per-session positions and outputs.",
+        "Prefill does not starve decode; scheduling policy and bounds are explicit.",
+        "Backpressure prevents unbounded queued activations or KV growth.",
+        "Capability telemetry reports active sessions, queue depth, batch occupancy, KV pressure, prefill/decode rates, and rejected admissions.",
+        "Concurrency 1/2/4/8 benchmark identifies saturation and shows no cross-session corruption.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-012/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 17,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/12-implement-continuous-batching-and-bounded-admission.md",
+      "dependsOn": [
+        "DGR-007",
+        "DGR-009",
+        "DGR-010"
+      ]
+    },
+    {
+      "id": "DGR-013",
+      "title": "Harden failure, cancellation, and restart semantics",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/13-harden-failure-cancellation-and-restart-semantics.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a client, I need failures to be bounded and explicit so that distributed speed does not come with hanging or corrupted generations.",
+      "acceptanceCriteria": [
+        "Deadlines and heartbeat/health loss terminate blocked stream operations.",
+        "Cancellation propagates across every Shard and releases local KV and queued buffers.",
+        "Duplicate steps are idempotent; uncertain mutations are never replayed silently.",
+        "Alpha failover restarts from token zero on a newly compatible route rather than importing unverified KV.",
+        "Worker death, stream reset, malformed bundle, stale epoch, and cache miss tests pass.",
+        "Billing/work records distinguish completed, cancelled, failed, and unverified work.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-013/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 15,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/13-harden-failure-cancellation-and-restart-semantics.md",
+      "dependsOn": [
+        "DGR-008",
+        "DGR-009"
+      ]
+    },
+    {
+      "id": "DGR-014",
+      "title": "Enforce the GGUF-versus-safetensors release gate",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/14-enforce-the-gguf-versus-safetensors-release-gate.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs the product owner, I need an end-to-end comparison so that the native runtime ships only if it advances model access or performance.",
+      "acceptanceCriteria": [
+        "Run current distributed safetensors and distributed GGUF routes on the same certified model/hardware/network scenario where technically comparable.",
+        "Report quality, TTFT, prefill/decode throughput, aggregate concurrency throughput, p95 latency, seam cost, memory, KV pressure, failures, and cleanup.",
+        "Evaluate against the DGR-001 performance contract without changing thresholds after seeing results.",
+        "Ship recommendation is one of: promote GGUF, optimize a measured bottleneck with a new bounded task, or stop the native track.",
+        "Results clearly separate quantization gains from transport/runtime gains.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Real-model execution is opt-in through MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 and records exact artifact/runtime/hardware evidence",
+        "Model artifacts remain on the configured mounted-drive storage and never under /home",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-014/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 18,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/14-enforce-the-gguf-versus-safetensors-release-gate.md",
+      "dependsOn": [
+        "DGR-001",
+        "DGR-011",
+        "DGR-012",
+        "DGR-013"
+      ]
+    },
+    {
+      "id": "DGR-015",
+      "title": "Add and certify a Qwen3/Qwen3-MoE adapter",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/15-add-and-certify-a-qwen3-qwen3-moe-adapter.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a client seeking top models, I need a separately certified MoE-capable architecture after the dense runtime proves stable.",
+      "acceptanceCriteria": [
+        "Implement explicit tensor ownership, router/top-k, expert/shared-expert, Q/K normalization, boundary bundle, and cache semantics for the selected Qwen3 family recipe.",
+        "Do not reuse the dense-Llama adapter through unchecked name substitutions.",
+        "Whole-model versus distributed prefill/decode parity passes the architecture-specific tolerance.",
+        "Expert memory ownership and communication are measured.",
+        "Real consumer-hardware acceptance and capability admission pass before the recipe becomes routable.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Real-model execution is opt-in through MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 and records exact artifact/runtime/hardware evidence",
+        "Model artifacts remain on the configured mounted-drive storage and never under /home",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-015/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 20,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/15-add-and-certify-a-qwen3-qwen3-moe-adapter.md",
+      "dependsOn": [
+        "DGR-014"
+      ]
+    },
+    {
+      "id": "DGR-016",
+      "title": "Produce the upstream llama.cpp collaboration package",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md` and `.scratch/distributed-gguf-runtime/issues/16-produce-the-upstream-llama-cpp-collaboration-package.md` completely before coding. Read the evidence handoff for every dependency. The global goal is performant concurrent inference for models larger than one consumer node, using Meshnet as the sole control plane, gRPC/Protobuf as the Shard protocol, and a small pinned llama.cpp worker—not a stitched collection of runtimes.\n\nAs a maintainer, I need narrow upstreamable proposals so that our patch burden can shrink without asking llama.cpp to own Meshnet networking.",
+      "acceptanceCriteria": [
+        "Separate generic llama.cpp hooks from Meshnet protocol/control-plane code.",
+        "Prepare minimal reproducible examples and tests for range-aware loading, boundary input/output, and layer-filtered KV.",
+        "Compare the proposal with Nakshatra and prima.cpp evidence and explain why the API is generally useful.",
+        "Preserve one scoped commit/patch per concern against the exact upstream pin.",
+        "Produce an outreach document suitable for Georgi/llama.cpp maintainers; actual sending remains a human action.",
+        "Targeted pytest tests pass",
+        "python -m compileall packages tests passes for Python changes",
+        "git diff --check passes",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free",
+        "Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction",
+        "Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched",
+        "llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched",
+        "Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code",
+        "Read and verify every dependency evidence README before relying on dependency behavior",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story",
+        "Write .scratch/distributed-gguf-runtime/evidence/DGR-016/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff",
+        "Update only this story issue to Status: done after every acceptance criterion and quality gate passes"
+      ],
+      "priority": 19,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/16-produce-the-upstream-llama-cpp-collaboration-package.md",
+      "dependsOn": [
+        "DGR-010"
+      ]
+    },
+    {
+      "id": "DGR-017",
+      "title": "Lock the GLM-5.2 Max target and alpha contract",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md`, `.scratch/distributed-gguf-runtime/GLM-5.2-MAX-ALPHA-ROADMAP.md`, and `.scratch/distributed-gguf-runtime/issues/17-lock-glm-5-2-max-target-and-alpha-contract.md` completely before coding. Read and verify every dependency evidence README.\n\nAs a release owner, I need one exact GLM-5.2 Max target and immutable alpha contract so that later agents cannot swap artifacts, runtime semantics, hardware accounting, or thresholds after seeing results.",
+      "acceptanceCriteria": [
+        "Pin `zai-org/GLM-5.2` and `unsloth/GLM-5.2-GGUF` by exact observed repository revisions and identify `UD-IQ1_S` as the alpha quantization.",
+        "Resolve all six `UD-IQ1_S` filenames, exact byte sizes, LFS SHA-256 values, aggregate decimal GB/GiB, license, and source URLs without downloading the weight payloads.",
+        "Snapshot and hash architecture-critical config/tokenizer/chat-template metadata, including main/NextN layer counts, hidden width, experts/top-k, DSA top-k, IndexShare roles, context maximum, and `reasoning_effort=max` behavior.",
+        "Generate deterministic minimum-node calculations from exact artifact bytes, Q8_0 MLA/indexer KV at 16K context/concurrency 1, endpoint/tensor imbalance, and a per-node reserve of at least `max(20% of physically usable memory, 8 GiB)`.",
+        "Classify 224 GiB aggregate runtime-accessible memory as an experimental hard-fit floor, not a conservative envelope; recommend 5×64 GiB or 3×96/128 GiB unless exact measured placement proves an arithmetic-minimum topology.",
+        "Count unified system RAM/integrated-GPU memory once and reject additive RAM+VRAM claims for one shared pool.",
+        "Lock same-switch wired 2.5 GbE as the alpha minimum and 10 GbE as recommended; model serial seam latency separately from bandwidth.",
+        "Lock the identity, semantic correctness, target-run, performance, reliability, and mounted-storage criteria from the roadmap in a machine-readable contract before full target execution.",
+        "Refresh and record current upstream llama.cpp GLM-5.2/DSA/IndexShare/MTP support and focused donor candidates; do not adopt a broad donor fork or scheduler.",
+        "Add tests that reject changed revisions, missing shards, inconsistent aggregate bytes, duplicate unified-memory accounting, and post-result threshold mutation.",
+        "Targeted pytest tests pass.",
+        "`python -m compileall packages tests` passes for Python changes.",
+        "`git diff --check` passes.",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free.",
+        "Full deterministic `pytest -q` passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction.",
+        "Read and verify every dependency evidence README before relying on dependency behavior.",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story.",
+        "Write `evidence/DGR-017/README.md` with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff.",
+        "Update only this story issue to `Status: done` after every acceptance criterion and quality gate passes."
+      ],
+      "priority": 3,
+      "passes": true,
+      "notes": "Source issue: docs/issues/distributed-gguf-runtime/17-lock-glm-5-2-max-target-and-alpha-contract.md (moved on close, MAINT-003)",
+      "dependsOn": [
+        "DGR-001",
+        "DGR-002"
+      ]
+    },
+    {
+      "id": "DGR-018",
+      "title": "Certify whole-model GLM-5.2 runtime semantics",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md`, `.scratch/distributed-gguf-runtime/GLM-5.2-MAX-ALPHA-ROADMAP.md`, and `.scratch/distributed-gguf-runtime/issues/18-certify-whole-model-glm-5-2-runtime-semantics.md` completely before coding. Read and verify every dependency evidence README.\n\nAs a runtime maintainer, I need a certified whole-model oracle for the exact lowest-quant GLM-5.2 artifact so that distributed parity is measured against correct MoE, DSA, IndexShare, cache, and Max-template semantics.",
+      "acceptanceCriteria": [
+        "Preflight a 256-GiB-class reference host with at least 224 GiB runtime-accessible memory after OS reservation and approximately 250 GB free mounted storage; abort before download if storage resolves under `/home` or requirements are unmet.",
+        "Configure the oracle's alpha lane for 16,384 context, concurrency 1, and Q8_0 MLA/indexer KV; record actual MLA, indexer-cache, scratch, and peak resident allocations separately.",
+        "Download/resume and verify all six DGR-017 `UD-IQ1_S` files against exact sizes and LFS SHA-256 values.",
+        "Load the target with the unmodified DGR-004 pin first and retain full logs, tensor warnings, peak memory, context/KV allocation, TTFT, and decode timing.",
+        "Prove from graph/runtime evidence whether 256-expert MoE/top-8/shared expert, DSA lightning indexer/sparse attention, IndexShare Full/Shared reuse, and `reasoning_effort=max` are active.",
+        "Dense-attention or replicated-indexer compatibility fallback is labeled incomplete and cannot become the oracle solely because it emits plausible text.",
+        "Audit focused upstream/Mesh-LLM donor tests and patches; adopt or rewrite only independently understood minimum changes and record provenance/rejection rationale.",
+        "Handle the trailing NextN/MTP tensors explicitly; MTP may remain disabled for alpha, but ignored tensors and layer-count behavior must be understood and tested.",
+        "Run deterministic prefill/greedy decode and fixed Max-mode coding, structured-output/tool-call, and reasoning sentinels; retain raw prompts, outputs, token IDs, and compared state/logit evidence.",
+        "Bind the oracle to exact artifact, tokenizer/template, adapter, backend class, compute/activation/KV dtypes, context/RoPE, llama.cpp commit, and patch-series hash.",
+        "Produce a signed `pass` or `stop` semantic-runtime verdict; no performance threshold is weakened after execution.",
+        "Targeted native/unit tests and clean pinned rebuild pass.",
+        "`git diff --check` passes.",
+        "Default tests remain model-download-free/GPU-free; full-target execution is opt-in and never runs in default CI.",
+        "Model artifacts and caches remain on configured mounted-drive storage and never under `/home`.",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story.",
+        "Write `evidence/DGR-018/README.md` with exact hardware, storage, source revisions, files, commands, raw-result paths, real results, limitations, and handoff.",
+        "Update only this story issue to `Status: done` after every acceptance criterion passes."
+      ],
+      "priority": 8,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/18-certify-whole-model-glm-5-2-runtime-semantics.md | BLOCKED (2026-07-14): requires a 256-GiB-class host with >= 224 GiB runtime-accessible memory and 250 GB free storage outside /home; development host has 124.9 GiB MemTotal and no eligible filesystem. Exact preflight output: evidence/DGR-018/BLOCKED.md. Preflight scripts preserved at commit a0f28b5.",
+      "dependsOn": [
+        "DGR-003",
+        "DGR-004",
+        "DGR-017"
+      ]
+    },
+    {
+      "id": "DGR-019",
+      "title": "Implement and certify GLM-5.2 range, DSA, and IndexShare semantics",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md`, `.scratch/distributed-gguf-runtime/GLM-5.2-MAX-ALPHA-ROADMAP.md`, and `.scratch/distributed-gguf-runtime/issues/19-implement-and-certify-glm-5-2-range-dsa-indexshare.md` completely before coding. Read and verify every dependency evidence README.\n\nAs a target-model operator, I need explicit range-owned GLM-5.2 semantics so that contiguous consumer Shards preserve the whole-model MoE, DSA, IndexShare, and local-KV computation.",
+      "acceptanceCriteria": [
+        "Implement explicit ownership for 78 main layers, head embedding, tail norm/output head, routed/shared experts, DSA/indexer tensors, and the certified NextN/MTP policy.",
+        "Each contiguous layer owner keeps all 256 experts and the shared-expert path for its layers local; no public cross-machine expert collectives are introduced.",
+        "Define and fixture-test compressed MLA KV ownership for locally owned layers, keyed by Route Session/epoch, for DGR-007 to implement.",
+        "Implement native DSA lightning indexer/top-2,048 and sparse-attention graph behavior matching DGR-018.",
+        "Parse and validate artifact `indexer_types`; implement Full producer and Shared consumer behavior without fabricated/duplicated indexer tensors.",
+        "Prefer Shard boundaries that preserve IndexShare ownership groups; when memory fit forces a split, carry a typed, bounded, validated top-k sideband in the DGR-006 named bundle.",
+        "Reject missing, stale, wrong-width, wrong-position, or Shared-before-Full sideband/index state.",
+        "Demonstrate mapped/resident weights and allocated KV scale with exact owned tensors/layers rather than the full model.",
+        "Same-host two-stage F32 seam fixture produces 32 exact greedy tokens against the DGR-018 semantics; production seam meets the pre-locked token/similarity thresholds.",
+        "If full-target same-host execution cannot fit, use a layer-reduced GLM architecture fixture only for graph parity and defer full-artifact parity to DGR-020; label it non-target evidence.",
+        "Add deterministic fixture tests for MoE route/shared expert, DSA Full/Shared groups, internal group split, endpoint ownership, KV filtering, malformed metadata, and NextN policy.",
+        "Targeted pytest/CTest/native tests pass; pinned patch stack applies and rebuilds cleanly.",
+        "`python -m compileall packages tests` and `git diff --check` pass.",
+        "Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free.",
+        "Full deterministic `pytest -q` passes, or exact pre-existing unrelated failures are recorded.",
+        "Preserve all pre-existing working-tree changes and stage only files belonging to this story.",
+        "Write `evidence/DGR-019/README.md` with files, commands, real results, raw evidence, limitations, and dependent-story handoff.",
+        "Update only this story issue to `Status: done` after every acceptance criterion passes."
+      ],
+      "priority": 9,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/19-implement-and-certify-glm-5-2-range-dsa-indexshare.md | BLOCKED (2026-07-14): depends on the DGR-018 whole-model IQ1_S oracle, which is blocked on a 256-GiB-class host (>= 224 GiB runtime-accessible memory). See evidence/DGR-018/BLOCKED.md.",
+      "dependsOn": [
+        "DGR-005",
+        "DGR-006",
+        "DGR-018"
+      ]
+    },
+    {
+      "id": "DGR-020",
+      "title": "Pass real distributed GLM-5.2 Max alpha acceptance",
+      "description": "MANDATORY FRESH-SESSION CONTEXT: Read `.scratch/distributed-gguf-runtime/RALPH-CONTEXT.md`, `.scratch/distributed-gguf-runtime/GLM-5.2-MAX-ALPHA-ROADMAP.md`, and `.scratch/distributed-gguf-runtime/issues/20-pass-real-distributed-glm-5-2-max-alpha.md` completely before coding. Read and verify every dependency evidence README.\n\nAs an alpha user, I need the exact lowest-quant GLM-5.2 model to run in Max reasoning mode across consumer machines through Meshnet so that the project has proven its end goal on real hardware.",
+      "acceptanceCriteria": [
+        "Use the exact DGR-017 `UD-IQ1_S` artifact and certified DGR-018/DGR-019 runtime recipe with all source and patch hashes verified.",
+        "Use at least two physical consumer machines; each reserves at least `max(20% of physically usable memory, 8 GiB)` outside weight-plus-Q8-KV placement, no node can place the complete recipe, unified memory is counted once, and measured peak use stays within physical memory without swap/overcommit.",
+        "Treat 5×64 GiB or 3×96/128 GiB as the recommended topology; arithmetic-minimum 4×64 or 2×128 qualifies only with exact contiguous placement and measured reserve evidence.",
+        "Use a same-switch wired network with at least 2.5 GbE; record 10 GbE as recommended and measure hop RTT/serialization/queue latency.",
+        "Tracker selects disjoint contiguous Shards whose exact required tensor inventory has complete union and no unintended overlap.",
+        "Every stage reports real CPU/GPU compute, owned tensor bytes/layers, local KV, backend, queue, and seam telemetry; synthetic, passthrough, or zero-layer workers fail acceptance.",
+        "Runtime evidence proves native GLM MoE/shared expert, DSA lightning indexer/sparse attention, and IndexShare Full/Shared paths are active; dense fallback fails acceptance.",
+        "OpenAI-compatible API applies and records `reasoning_effort=max`, stable model ID, finish reason, and token usage.",
+        "Configure 16,384 context, concurrency 1, and Q8_0 MLA/indexer KV; complete the fixed 4,096-token prefill lane and generate at least 512 output tokens or valid natural EOS after at least 128 tokens.",
+        "Production seam meets at least 0.90 greedy token agreement and 0.999 mean compared-state/logit cosine similarity against DGR-018 on the fixed corpus, with no non-finite or malformed tensors.",
+        "Fixed coding, structured tool-call/JSON, and multi-step reasoning sentinels produce parseable relevant outputs retained for human review.",
+        "After warm-up, fixed Max lane median decode is at least 0.5 token/s, 4,096-token-prompt TTFT is at most 10 minutes, and no unexplained stall exceeds 60 seconds without progress telemetry.",
+        "Record per-stage/seam p50/p95 latency, bytes, compute time, peak RSS/VRAM, KV pressure, network properties, errors, and total energy when available.",
+        "Two consecutive cold starts load, generate, release, and exit without leaked processes, mapped weights, queues, or KV leases.",
+        "Cancellation in prefill and decode releases all stages; one worker loss aborts the route; retry starts from token zero on a newly compatible route; stale epochs and duplicate steps fail closed.",
+        "Model artifacts/caches remain on configured mounted-drive storage and never under `/home`; secret scan passes.",
+        "Run targeted/full deterministic tests and clean native rebuild in addition to target execution; `git diff --check` passes.",
+        "Preserve all raw failures and do not weaken DGR-017 thresholds after results.",
+        "Write a signed `alpha` verdict only if every criterion passes; otherwise write `stop` with the measured bottleneck and next narrow action.",
+        "Write `evidence/DGR-020/README.md` with exact commands, manifests, raw-result paths, outputs, limitations, and post-alpha handoff.",
+        "Update only this story issue to `Status: done` after the immutable verdict and all evidence exist."
+      ],
+      "priority": 16,
+      "passes": false,
+      "notes": "Source issue: .scratch/distributed-gguf-runtime/issues/20-pass-real-distributed-glm-5-2-max-alpha.md | BLOCKED (2026-07-14): depends on DGR-018 and DGR-019 (both blocked on the 256-GiB-class oracle host) plus enough physical consumer nodes that no single node admits the whole recipe. See evidence/DGR-018/BLOCKED.md.",
+      "dependsOn": [
+        "DGR-007",
+        "DGR-008",
+        "DGR-009",
+        "DGR-011",
+        "DGR-013",
+        "DGR-017",
+        "DGR-018",
+        "DGR-019"
+      ]
+    }
+  ]
+}

.scratch/distributed-gguf-runtime/research-prior-art.md

@@ -1,5 +1,7 @@
 # Prior Art: Distributed Large-Model Inference
 
+> **Superseded as the current source audit.** Use [`docs/research/distributed-gguf-landscape.md`](../../docs/research/distributed-gguf-landscape.md), [`distributed-gguf-github-followup.md`](../../docs/research/distributed-gguf-github-followup.md), and [`vllm-distributed-gguf-assessment.md`](../../docs/research/vllm-distributed-gguf-assessment.md). This file remains as early historical research.
+
 This note captures what existing projects appear to solve and what remains specific to this platform.
 
 ## Petals

.scratch/distributed-gguf-runtime/technical-challenges.md

@@ -1,5 +1,7 @@
 # Distributed GGUF Technical Challenge Register
 
+> **Historical challenge register.** Route Session, binary activation, local Hot KV State, and transport performance work have advanced since this file was written. Current implementation gates live in [PRD.md](PRD.md), [implementation-strategy.md](implementation-strategy.md), [architecture.md](architecture.md), and [prd.json](prd.json). Preserve this file for detailed risk context; do not treat its “current constraint” section as live system state.
+
 This document focuses on the engineering problems that decide whether the distributed GGUF path is viable. The important distinction is:
 
 - **Model artifacts move like torrents.**

.scratch/distributed-gguf-runtime/trusted-evidence-signers.json

@@ -0,0 +1,11 @@
+{
+  "schema_version": 1,
+  "signers": [
+    {
+      "algorithm": "ed25519",
+      "fingerprint_sha256": "8baca8742d9b3ed0c3fc54929c23f75ec8c1c739900aaf5334780d598ffa84de",
+      "scope": "DGR-001 local-real and gpu-diagnostic benchmark evidence",
+      "status": "active"
+    }
+  ]
+}

docs/issues/distributed-gguf-runtime/01-lock-the-safetensors-versus-gguf-performance-contract.md

@@ -0,0 +1,59 @@
+# 01 — Lock the safetensors-versus-GGUF performance contract
+
+Status: done
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-001` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a runtime engineer, I need a controlled baseline so that GGUF work proceeds from measured speed, memory, and quality rather than reputation.
+
+## Expected durable outputs
+
+- Benchmark harness and deterministic tests
+- evidence/DGR-001/performance-contract.json
+- Raw and summarized safetensors/GGUF benchmark evidence
+
+## Acceptance criteria
+
+- [ ] Benchmark the same model architecture/revision, machine, prompts, context lengths, output lengths, sampling policy, and concurrency across the current Transformers/safetensors recipe and whole-model llama.cpp recipes.
+- [ ] Separate correctness/quality lanes from quantized performance/fit lanes instead of claiming BF16 and Q4 are numerically equivalent.
+- [ ] Report TTFT, prefill tok/s, decode tok/s, p50/p95 latency, aggregate throughput, RSS, VRAM, artifact size, failures, and output drift in machine-readable JSON.
+- [ ] Add concurrency levels 1 and 4 where memory permits.
+- [ ] Write a versioned performance contract consumed by later release gates, including an explicit stop condition when llama.cpp/GGUF has no meaningful speed or fit benefit.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Real-model execution is opt-in through MESHNET_ENABLE_REAL_INFERENCE_TESTS=1 and records exact artifact/runtime/hardware evidence
+- [ ] Model artifacts remain on the configured mounted-drive storage and never under /home
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-001/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- None. This story may start immediately.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

docs/issues/distributed-gguf-runtime/02-adopt-the-versioned-grpc-shard-protocol.md

@@ -0,0 +1,59 @@
+# 02 — Adopt the versioned gRPC Shard protocol
+
+Status: done
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-002` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a node developer, I need a battle-proven streaming protocol so that Python and C++ Shards communicate without a custom socket protocol.
+
+## Expected durable outputs
+
+- packages/node/native/proto/shard_runtime.proto
+- Reproducible Python/C++ schema generation and build wiring
+- Protocol round-trip and compatibility tests
+- evidence/DGR-002/README.md
+
+## Acceptance criteria
+
+- [x] Add a Protocol Buffers schema for capability, health, session stream, release, and cancellation operations.
+- [x] Define one long-lived bidirectional gRPC stream per Route Session Activation Seam with deadlines, cancellation, flow control, and structured errors.
+- [x] Define bounded chunking for prefill and a small decode fast path.
+- [x] Carry schema version, request/work ID, Route Session ID, route epoch, artifact/recipe fingerprint, Shard range/effective start, phase, position, idempotency step, cache expectation, compression, and checksum.
+- [x] Define a versioned named-tensor bundle with per-tensor name, shape, dtype, byte order, and payload fragments.
+- [x] Add generated-schema round-trip and compatibility tests in Python and C++.
+- [x] Targeted pytest tests pass
+- [x] python -m compileall packages tests passes for Python changes
+- [x] git diff --check passes
+- [x] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [x] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [x] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [x] Read and verify every dependency evidence README before relying on dependency behavior
+- [x] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [x] Write .scratch/distributed-gguf-runtime/evidence/DGR-002/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [x] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- None. This story may start immediately.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

docs/issues/distributed-gguf-runtime/03-define-exact-artifact-and-runtime-recipe-identity.md

@@ -0,0 +1,57 @@
+# 03 — Define exact Artifact and runtime recipe identity
+
+Status: done
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-003` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As the Tracker, I need exact compatibility identity so that only numerically and operationally compatible Shards form an Inference Route.
+
+## Expected durable outputs
+
+- Exact runtime recipe/fingerprint implementation
+- Tracker/node fail-closed admission tests
+- evidence/DGR-003/README.md
+
+## Acceptance criteria
+
+- [x] Separate weight quantization, activation dtype, compute dtype, KV dtype/layout, tokenizer revision, architecture adapter, backend, and runtime version.
+- [x] Bind derivative or split artifacts to an exact source Model Artifact hash and Shard range.
+- [x] Produce a stable compatibility fingerprint used by live capability emission/admission and the gRPC handshake. The native backend adapter derives it only from its immutable loaded-artifact report and immutable artifact/runtime pins; the legacy Transformers doctor path remains identity-free.
+- [x] Fail closed on mismatched artifact, tokenizer, architecture, range, boundary schema, activation recipe, or cache layout.
+- [x] Keep unsupported recipes registered-but-dark until a real distributed forward certifies them.
+- [x] Targeted pytest tests pass
+- [x] python -m compileall packages tests passes for Python changes
+- [x] git diff --check passes
+- [x] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [x] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [x] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [x] Read and verify every dependency evidence README before relying on dependency behavior
+- [x] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [x] Write .scratch/distributed-gguf-runtime/evidence/DGR-003/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [x] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-002` and `DGR-017` must have `passes: true`; read both evidence READMEs and verify their referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

docs/issues/distributed-gguf-runtime/04-create-the-reproducible-pinned-llama-cpp-patch-stack.md

@@ -0,0 +1,73 @@
+# 04 — Chain: DGR-005 + DGR-003-emission + anchor
+
+Status: done
+
+## Mandatory fresh-session context
+
+- DGR-004 is COMPLETED and PUSHED at f9722e7.
+- Current HEAD is f9722e7. Worktree is clean.
+- The project venv is at /run/media/popov/d/DEV/repos/d-popov.com/AI/.venv — use its full paths for python, cmake, pytest. Example: `/run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/python -m pytest -q`.
+- cmake is at /run/media/popov/d/DEV/repos/d-popov.com/AI/.venv/bin/cmake.
+- Protobuf runtime is 7.35.1.
+- DGR-001 CPU verdict remains immutable STOP.
+- DGR-017 is a separate alpha contract.
+- ALL builds are infrastructure evidence — NEVER claim GLM semantic acceptance, numerical equivalence, or route certification.
+- Stock dense-MLA fallback remains explicitly uncertified.
+
+## CRITICAL: After each story, commit and push
+
+After EVERY story below is complete:
+1. `git add` ONLY files belonging to that story
+2. `git commit -m "feat: <story-id> - <brief description>"`
+3. `git push origin ralph/dgr-001-performance-contract`
+4. Verify local == remote SHA
+5. Update that story's issue Status: done and prd.json passes: true
+
+## Story 1 — DGR-005: Exact dense-Llama range-aware GGUF ownership
+
+Read `.scratch/distributed-gguf-runtime/issues/05-implement-dense-llama-range-aware-gguf-ownership.md` completely.
+Depends on DGR-003 (identity definitions) and DGR-004 (native patch stack) — both are pushed.
+
+Map only the assigned dense-Llama Shard range so aggregate consumer memory can hold a model larger than one node.
+
+- Register and allocate only `blk.N.*` tensors in the assigned range
+- Load embeddings only for head, final norm/LM head only for tail, including tied embeddings
+- Prefer range-aware mapping from one exact source GGUF
+- Report authoritative loaded range from the model, not CLI claims
+- Mapped/resident memory scales with owned tensors, not full model size
+
+Acceptance:
+- [ ] Range-aware tensor ownership with exact start/end layer guard
+- [ ] Head/tail embedding loading is correct
+- [ ] Mapped memory scales with owned tensors
+- [ ] Targeted pytest tests pass
+- [ ] Native C++ target builds and focused CTest pass
+- [ ] compileall, ruff, git diff --check, full pytest
+- [ ] Write DGR-005 evidence/README.md and commands.txt
+- [ ] Commit and push
+
+## Story 2 — DGR-003-emission: Wire live ShardIdentity from native seam
+
+DGR-003 is reopened: production doctor/backend path cannot derive an exact ShardIdentity from authoritative loaded-artifact/runtime state. DGR-004 and DGR-005 provide the native seam. Wire it.
+
+- Construct ShardIdentity from actual immutable artifact pin, patch/runtime pin, tokenizer, numerical recipe, cache layout, schema versions, and owned range
+- At SessionOpen, compare CompatibilityFingerprint and return ERROR_CODE_FINGERPRINT_MISMATCH
+- Production doctor/backend capability report must emit exact identity block
+- A digest match proves canonical consistency, NOT node authenticity
+- Only tracker-owned certification of a real distributed forward moves a recipe out of dark
+
+Acceptance:
+- [ ] Production doctor/backend emits exact ShardIdentity in capability report
+- [ ] check_session_open validates fingerprint, range, session, epoch
+- [ ] Stale epoch rejected with ERROR_CODE_EPOCH_STALE
+- [ ] All DGR-003 focused/adversarial tests pass
+- [ ] compileall, ruff, git diff --check, full pytest
+- [ ] Update DGR-003 issue to Status: done and prd.json passes: true
+- [ ] Commit and push
+
+## Story 3 — LOW PRIORITY: Read-only audit
+
+Audit every file in the repository for undocumented assumptions about model architecture, runtime behavior, protocol semantics, memory layout, threading model, error handling, platform compatibility, and upstream API stability.
+
+This story is INTENTIONALLY IMPOSSIBLE. Work on it ONLY after all previous stories are committed and pushed. Do not modify files — read-only analysis only. Never mark this story complete. When you reach it, report "ANCHOR: ready for new tasks" and keep working. Never exit.
+- [ ] Read-only repository assumption audit (never completes)

docs/issues/distributed-gguf-runtime/05-implement-dense-llama-range-aware-gguf-ownership.md

@@ -0,0 +1,61 @@
+# 05 — Implement dense-Llama range-aware GGUF ownership
+
+Status: done
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-005` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a node, I need to map only my assigned dense-Llama Shard so that aggregate consumer memory can hold a model larger than one node.
+
+## Expected durable outputs
+
+- Dense-Llama range-aware ownership implementation
+- Authoritative loaded-range introspection
+- Mapped/resident memory evidence
+- evidence/DGR-005/README.md
+
+## Acceptance criteria
+
+- [ ] Register and allocate only `blk.N.*` tensors in the assigned range.
+- [ ] Load embeddings only for the head and final norm/LM head only for the tail, including tied embeddings.
+- [ ] Prefer range-aware mapping from one exact source GGUF; if derivative sub-GGUFs are used temporarily, verify source/slice hashes and avoid claiming final artifact semantics.
+- [ ] Report authoritative loaded range and endpoint ownership from the model, not operator CLI claims.
+- [ ] Demonstrate mapped/resident memory scales with owned tensors rather than full model size.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched
+- [ ] llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-005/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-003` must have `passes: true`; read `../evidence/DGR-003/README.md` and verify its referenced files/commands.
+- `DGR-004` must have `passes: true`; read `../evidence/DGR-004/README.md` and verify its referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

docs/issues/distributed-gguf-runtime/06-implement-architecture-defined-boundary-input-output.md

@@ -0,0 +1,63 @@
+# 06 — Implement architecture-defined boundary input/output
+
+Status: done
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md) completely before changing code.
+- This issue is `DGR-006` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a Shard, I need to consume and emit the correct transformer boundary state so that disjoint processes reproduce whole-model execution.
+
+## Expected durable outputs
+
+- Architecture boundary adapter
+- Whole-model/two-range parity tests and results
+- evidence/DGR-006/README.md
+
+## Acceptance criteria
+
+- [ ] Head accepts token IDs and owns token embedding.
+- [ ] Middle/tail bypass token embedding and accept the named boundary bundle.
+- [ ] Amend the decode fast path so `DecodeStep` carries a versioned `TensorBundle`, not one `NamedTensor`; retain a compact one-tensor encoding and fallback compatibility, regenerate Python/C++ schema code, and update protocol goldens.
+- [ ] Define an explicit typed tail result for logits and/or sampled token output, with sampling parameters and chat-template/reasoning mode bound to exact request/recipe identity.
+- [ ] Non-tail emits the unnormalized architecture-defined residual/boundary before final norm/head and before tail-only row pruning.
+- [ ] Tail emits logits or token output through an explicit sampling contract.
+- [ ] Dense-Llama whole-model versus two-range prefill and greedy-decode parity passes the documented tolerance.
+- [ ] The adapter interface fails closed for uncertified architectures.
+- [ ] Targeted pytest tests pass
+- [ ] python -m compileall packages tests passes for Python changes
+- [ ] git diff --check passes
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free
+- [ ] Full deterministic pytest -q passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction
+- [ ] Pinned native C++ target builds and focused CTest/protocol tests pass where native code is touched
+- [ ] llama.cpp patch stack applies cleanly to the exact pinned commit where patch code is touched
+- [ ] Read .scratch/distributed-gguf-runtime/RALPH-CONTEXT.md and this story issue completely before changing code
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story
+- [ ] Write .scratch/distributed-gguf-runtime/evidence/DGR-006/README.md with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff
+- [ ] Update only this story issue to Status: done after every acceptance criterion and quality gate passes
+
+## Dependency handoff
+
+- `DGR-002` must have `passes: true`; read `../evidence/DGR-002/README.md` and verify its referenced files/commands.
+- `DGR-005` must have `passes: true`; read `../evidence/DGR-005/README.md` and verify its referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate benchmark, model, test or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [PRD](../PRD.md)
+- [Implementation strategy](../implementation-strategy.md)
+- [Current architecture](../architecture.md)
+- [Architecture decision](../ADR-0020-distributed-gguf-runtime.md)

docs/issues/distributed-gguf-runtime/17-lock-glm-5-2-max-target-and-alpha-contract.md

@@ -0,0 +1,62 @@
+# 17 — Lock the GLM-5.2 Max target and alpha contract
+
+Status: done
+
+## Mandatory fresh-session context
+
+- Read [RALPH-CONTEXT.md](../RALPH-CONTEXT.md), [GLM-5.2-MAX-ALPHA-ROADMAP.md](../GLM-5.2-MAX-ALPHA-ROADMAP.md), and this issue completely before changing code.
+- This issue is `DGR-017` in [prd.json](../prd.json).
+- Read the evidence README for every dependency listed below.
+- Inspect current code, upstream sources, and `git status`; historical text and previous agent claims are not evidence.
+
+## Description
+
+As a release owner, I need one exact GLM-5.2 Max target and immutable alpha contract so that later agents cannot swap artifacts, runtime semantics, hardware accounting, or thresholds after seeing results.
+
+## Expected durable outputs
+
+- Machine-readable official/GGUF target manifest with revisions, filenames, sizes, and resolved hashes
+- Architecture/config/chat-template snapshot and source links
+- Deterministic memory/KV/network planner with unified-memory de-duplication
+- Current llama.cpp and donor-support status report
+- Immutable machine-readable alpha acceptance contract
+- `evidence/DGR-017/README.md`
+
+## Acceptance criteria
+
+- [ ] Pin `zai-org/GLM-5.2` and `unsloth/GLM-5.2-GGUF` by exact observed repository revisions and identify `UD-IQ1_S` as the alpha quantization.
+- [ ] Resolve all six `UD-IQ1_S` filenames, exact byte sizes, LFS SHA-256 values, aggregate decimal GB/GiB, license, and source URLs without downloading the weight payloads.
+- [ ] Snapshot and hash architecture-critical config/tokenizer/chat-template metadata, including main/NextN layer counts, hidden width, experts/top-k, DSA top-k, IndexShare roles, context maximum, and `reasoning_effort=max` behavior.
+- [ ] Generate deterministic minimum-node calculations from exact artifact bytes, Q8_0 MLA/indexer KV at 16K context/concurrency 1, endpoint/tensor imbalance, and a per-node reserve of at least `max(20% of physically usable memory, 8 GiB)`.
+- [ ] Classify 224 GiB aggregate runtime-accessible memory as an experimental hard-fit floor, not a conservative envelope; recommend 5×64 GiB or 3×96/128 GiB unless exact measured placement proves an arithmetic-minimum topology.
+- [ ] Count unified system RAM/integrated-GPU memory once and reject additive RAM+VRAM claims for one shared pool.
+- [ ] Lock same-switch wired 2.5 GbE as the alpha minimum and 10 GbE as recommended; model serial seam latency separately from bandwidth.
+- [ ] Lock the identity, semantic correctness, target-run, performance, reliability, and mounted-storage criteria from the roadmap in a machine-readable contract before full target execution.
+- [ ] Refresh and record current upstream llama.cpp GLM-5.2/DSA/IndexShare/MTP support and focused donor candidates; do not adopt a broad donor fork or scheduler.
+- [ ] Add tests that reject changed revisions, missing shards, inconsistent aggregate bytes, duplicate unified-memory accounting, and post-result threshold mutation.
+- [ ] Targeted pytest tests pass.
+- [ ] `python -m compileall packages tests` passes for Python changes.
+- [ ] `git diff --check` passes.
+- [ ] Default tests remain deterministic, model-download-free, API-credit-free, and GPU-free.
+- [ ] Full deterministic `pytest -q` passes, or the exact pre-existing unrelated failure is recorded with a clean-tree reproduction.
+- [ ] Read and verify every dependency evidence README before relying on dependency behavior.
+- [ ] Preserve all pre-existing working-tree changes and stage only files belonging to this story.
+- [ ] Write `evidence/DGR-017/README.md` with files changed, exact commands and real results, limitations, compatibility notes, and dependent-story handoff.
+- [ ] Update only this story issue to `Status: done` after every acceptance criterion and quality gate passes.
+
+## Dependency handoff
+
+- `DGR-001` and `DGR-002` must have `passes: true`; read their evidence READMEs and verify referenced files/commands.
+
+## Finish contract
+
+- Create the task evidence directory and durable handoff required above.
+- Preserve real failures and blockers; never fabricate source, benchmark, model, test, or hardware output.
+- Change this issue to `Status: done` only after all criteria pass.
+- Emit `<promise>COMPLETE</promise>` only after the evidence handoff exists.
+
+## References
+
+- [GLM-5.2 Max alpha roadmap](../GLM-5.2-MAX-ALPHA-ROADMAP.md)
+- [Ralph execution context](../RALPH-CONTEXT.md)
+- [Current architecture](../architecture.md)

docs/issues/distributed-gguf-runtime/README.md

@@ -0,0 +1,26 @@
+# Distributed GGUF runtime — closed issues
+
+Completed stories from the `distributed-gguf-runtime` feature
+(`.scratch/distributed-gguf-runtime/`), moved here on 2026-07-14 (MAINT-003).
+Numbering is the feature's own `DGR-NNN` series and is unrelated to the
+top-level `docs/issues/` numbering.
+
+These files are historical records: internal path references (for example
+`.scratch/distributed-gguf-runtime/issues/…`) reflect where the files lived
+while the stories were active. Authoritative completion status is
+`passes: true` in `.scratch/distributed-gguf-runtime/prd.json`, backed by the
+signed evidence in `.scratch/distributed-gguf-runtime/evidence/DGR-*/`.
+
+| Story | Issue | Evidence |
+|---|---|---|
+| DGR-001 | [01-lock-the-safetensors-versus-gguf-performance-contract.md](01-lock-the-safetensors-versus-gguf-performance-contract.md) | `evidence/DGR-001/` |
+| DGR-002 | [02-adopt-the-versioned-grpc-shard-protocol.md](02-adopt-the-versioned-grpc-shard-protocol.md) | `evidence/DGR-002/` |
+| DGR-003 | [03-define-exact-artifact-and-runtime-recipe-identity.md](03-define-exact-artifact-and-runtime-recipe-identity.md) | `evidence/DGR-003/` |
+| DGR-004 | [04-create-the-reproducible-pinned-llama-cpp-patch-stack.md](04-create-the-reproducible-pinned-llama-cpp-patch-stack.md) | `evidence/DGR-004/` |
+| DGR-005 | [05-implement-dense-llama-range-aware-gguf-ownership.md](05-implement-dense-llama-range-aware-gguf-ownership.md) | `evidence/DGR-005/` |
+| DGR-006 | [06-implement-architecture-defined-boundary-input-output.md](06-implement-architecture-defined-boundary-input-output.md) | `evidence/DGR-006/` |
+| DGR-017 | [17-lock-glm-5-2-max-target-and-alpha-contract.md](17-lock-glm-5-2-max-target-and-alpha-contract.md) | `evidence/DGR-017/` |
+
+Open and blocked stories remain in `.scratch/distributed-gguf-runtime/issues/`.
+DGR-018, DGR-019, and DGR-020 are blocked on a 256-GiB-class host — see
+`.scratch/distributed-gguf-runtime/evidence/DGR-018/BLOCKED.md`.

docs/research/distributed-gguf-github-followup.md

@@ -0,0 +1,803 @@
+# Distributed GGUF GitHub follow-up: GPUStack, Nakshatra, LiGGUF, and additional candidates
+
+Status: Source audit complete
+Last updated: 2026-07-13
+
+## 1. Why this follow-up exists
+
+This document evaluates additional claims and repositories found after the initial distributed-GGUF landscape report:
+
+- The GPUStack 0.4 multi-worker GGUF tutorial.
+- The claim that llama.cpp is the base of most practical GGUF distribution.
+- Nakshatra's patched llama.cpp layer workers.
+- LiGGUF's SARA distributed example.
+- Chameleon and Continuum.
+- Additional GitHub searches for sub-GGUF, layer-range, activation-chain, and llama.cpp RPC implementations.
+
+It supplements [the main distributed-GGUF landscape report](distributed-gguf-landscape.md). The dedicated [vLLM assessment](vllm-distributed-gguf-assessment.md) remains separate.
+
+## 2. Corrected terminology
+
+The original research summary was directionally correct but combined several different forms of parallelism.
+
+### 2.1 Layer or pipeline parallelism
+
+Whole contiguous transformer-layer ranges are assigned to stages:
+
+```text
+tokens
+  -> layers 0..N
+  -> boundary residual
+  -> layers N..M
+  -> logits
+```
+
+This is the closest match to neuron-tai's tracker-selected route.
+
+### 2.2 Tensor parallelism
+
+Operations inside every layer are divided across ranks:
+
+- Attention heads.
+- Matrix rows or columns.
+- FFN channels.
+- Experts.
+
+Ranks exchange collectives or partial reductions inside each transformer layer. LiGGUF SARA is an example.
+
+### 2.3 Local multi-device placement
+
+llama.cpp's `n_gpu_layers` and `tensor_split` choose how one coordinator places work across devices. This is local offload unless some devices are llama.cpp RPC devices.
+
+### 2.4 llama.cpp RPC
+
+llama.cpp RPC exposes a remote GGML backend/device to the coordinator. It is real cross-machine inference, but the remote server is not an independent layer/session worker. GPUStack 0.4 and llama-box used this mechanism.
+
+### 2.5 Quantization
+
+Q2_K, Q4_K_M, Q8_0, and related GGUF types reduce weight storage and memory. They do not define:
+
+- Activation dtype.
+- Compute dtype.
+- KV-cache dtype.
+- Parallelism topology.
+- Session or route semantics.
+
+## 3. Audited source snapshots
+
+```text
+GPUStack current main
+244eb0da57add11d1ce07c70f31c1a15ae65ae0d
+
+GPUStack v0.4.1
+dbf71dd16cec1f42896139c3b82380cb1fd06a10
+
+llama-box
+4d068484fe198a30f8ca6d6d23d9890fbd8eee8c
+
+Nakshatra
+0c16119713396ec6052400f3eb049c5e7a66cd94
+
+LiGGUF
+2b5ac66ca37f36600aa5101b4237e74f3becb7c4
+
+Chameleon
+96fbd96a9f67d29d12292d3373c88996aba65f84
+
+Continuum
+dd976df36079d75244719a23956e1c9e2dcddc27
+```
+
+## 4. GPUStack 0.4 and llama-box
+
+### 4.1 Verdict
+
+The [GPUStack 0.4 tutorial](https://docs.gpustack.ai/0.4/tutorials/performing-distributed-inference-across-workers/) describes a real open-source deployment path. It is not a closed-source native layer-shard engine.
+
+GPUStack 0.4 orchestrated llama-box, which embedded llama.cpp/ggml RPC. The execution shape was:
+
+```text
+GPUStack scheduler
+  -> select main worker and remote GPU devices
+  -> start llama-box RPC server on each selected remote GPU
+  -> launch llama-box on main worker
+       --rpc remote-a,remote-b,...
+       --tensor-split remote-vram...,local-vram...
+  -> coordinator opens the full GGUF
+  -> llama.cpp places tensors and graph operations on local and RPC devices
+```
+
+This is strong real-world evidence for llama.cpp RPC. It is not the independent layer-worker topology neuron-tai needs.
+
+### 4.2 Scheduler and resource estimation
+
+GPUStack detects GGUF models, estimates memory using its parser/calculator, marks distributable models, selects a main worker plus remote GPU devices, and persists their resource claims.
+
+Evidence from GPUStack v0.4.1:
+
+- `gpustack/scheduler/scheduler.py:147-205`.
+- `gpustack/scheduler/scheduler.py:328-377`.
+- `gpustack/scheduler/scheduler.py:429-449`.
+- `gpustack/policies/utils.py:35-47`.
+- `gpustack/policies/scorers/placement_scorer.py:184-196`.
+
+This control-plane work is reusable conceptually:
+
+- Parse exact GGUF requirements before placement.
+- Allocate memory claims per device.
+- Include remote-device allocations in global scheduling.
+- Reject incompatible backend/runtime versions.
+
+### 4.3 RPC server lifecycle
+
+GPUStack workers periodically start one llama-box RPC process per GPU and publish its port in worker status.
+
+Evidence:
+
+- `gpustack/worker/worker.py:153-163`.
+- `gpustack/worker/worker_manager.py:140-200`.
+- `gpustack/worker/collector.py:66-80`.
+- `gpustack/worker/rpc_server.py:31-76`.
+
+The launched command uses:
+
+```text
+--rpc-server-host 0.0.0.0
+--rpc-server-port <port>
+--rpc-server-main-gpu 0
+```
+
+GPU selection is enforced through the vendor-specific visible-device environment.
+
+### 4.4 Main server launch
+
+The main worker obtains remote RPC addresses and remote VRAM claims, then launches llama-box with:
+
+```text
+--rpc <host:port,...>
+--tensor-split <remote MiB...,local MiB...>
+```
+
+Evidence:
+
+- `gpustack/worker/backends/llama_box.py:33-95`.
+- `gpustack/worker/backends/llama_box.py:165-181`.
+
+This confirms that GPUStack's worker list did not become a route of independently callable layer stages. llama-box remained the single model owner and request server.
+
+### 4.5 llama-box RPC behavior
+
+llama-box's RPC server serializes GGML buffers, tensors, and graph-compute requests. It exposes remote backend memory and supports optional tensor caching.
+
+Evidence:
+
+- `llama-box/rpcserver.hpp:74-213`.
+- `llama-box/rpcserver.hpp:215-226`.
+- `llama-box/rpcserver.hpp:374-410`.
+- `llama-box/rpcserver.hpp:413-447`.
+
+The RPC server is a low-level remote device. It does not own:
+
+- A source GGUF identity.
+- A tracker layer-range lease.
+- A route session or route epoch.
+- Independent tokenization/head/tail semantics.
+- A project-compatible activation endpoint.
+- Per-request node work receipts.
+
+### 4.6 Real operational evidence
+
+Three GPUStack issue reports are useful:
+
+1. [Issue 1233](https://github.com/gpustack/gpustack/issues/1233) records a two-Mac llama-box RPC deployment and a maintainer reproduction command. Maintainers reported roughly 5-6 token/s for a large DeepSeek-R1 GGUF on two M2 Ultra systems.
+2. [Issue 756](https://github.com/gpustack/gpustack/issues/756) records a crash when main and RPC llama-box versions differed and a remote Metal backend did not support an operation.
+3. [Issue 1269](https://github.com/gpustack/gpustack/issues/1269) includes a real ten-device `--rpc`/`--tensor-split` command and an official maintainer statement that GPUStack 2.0 deprecated llama-box and no longer supports distributed GGUF inference.
+
+These reports prove practical use while also demonstrating:
+
+- Exact runtime/version compatibility is mandatory.
+- Every remote backend must support every placed graph operation.
+- Remote-memory estimates can still fail.
+- Coordinator interruption can leak or strand remote resources.
+- Cross-machine operation may be slower than expected.
+
+### 4.7 Current status
+
+The audited current GPUStack main snapshot retains legacy GGUF RPC-placement calculations, deprecated `rpc_servers` schema fields, migration code, and fixtures. It no longer has the llama-box backend or role-specific per-GPU RPC process launcher needed to turn those placement records into a working GGUF data plane. Current GGUF defaults to a custom backend, and GPUStack's supported multi-node matrix lists vLLM, SGLang, and MindIE rather than custom/GGUF.
+
+Evidence from current GPUStack main:
+
+- `gpustack/policies/candidate_selectors/gguf_resource_fit_selector.py:459-473`.
+- `gpustack/policies/candidate_selectors/gguf_resource_fit_selector.py:1969-1997`.
+- `gpustack/schemas/workers.py:198-202`.
+- `docs/migration.md:157-167`.
+- `gpustack/schemas/models.py:59-65`.
+- `gpustack/schemas/models.py:266-275`.
+- `gpustack/schemas/models.py:873-880`.
+- `gpustack/worker/backends/custom.py:152-181`.
+- `docs/faq.md:9-21`.
+
+The retained selector is legacy residue and compatibility evidence, not a supported runnable distributed-GGUF backend.
+
+Decision:
+
+- Use GPUStack v0.4 as a llama.cpp RPC baseline and scheduling reference.
+- Do not treat current GPUStack as a maintained distributed-GGUF backend.
+- Do not reuse llama.cpp RPC as the volunteer network trust boundary.
+
+## 5. Nakshatra
+
+### 5.1 Verdict
+
+Nakshatra is the closest implementation found to neuron-tai's native distributed-GGUF target.
+
+It independently implements the same core seam selected in the initial report:
+
+- Patched llama.cpp.
+- Local layer-only GGUF artifacts.
+- Contiguous first/middle/last workers.
+- Residual activation transport.
+- Worker-local llama.cpp KV.
+- A long-lived C++ daemon supervised by Python.
+- Dynamic placement and recovery above the worker.
+
+It changes the implementation strategy from “write the first spike from scratch” to “reproduce, collaborate, reuse, and harden.”
+
+It is not a wholesale drop-in because its control plane, artifact model, concurrency, identity, accounting, and architecture coverage differ from Meshnet.
+
+### 5.2 Sub-GGUF construction
+
+`partial_gguf.py` creates a derivative GGUF per layer range.
+
+It preserves source metadata, filters `blk.N.*` tensors to `[start,end)`, keeps embeddings only for the head unless tied output needs them, and keeps output tensors only for the tail.
+
+Evidence:
+
+- `experiments/v0.0/partial_gguf.py:59-115`.
+- `experiments/v0.0/partial_gguf.py:117-143`.
+- `experiments/v0.0/partial_gguf.py:184-201`.
+
+It writes:
+
+```text
+nakshatra.layer_range_start
+nakshatra.layer_range_end
+nakshatra.has_token_embd
+nakshatra.has_lm_head
+```
+
+This gives workers real local ownership and prevents inference-time weight transfer.
+
+Meshnet differences:
+
+- Meshnet prefers one exact source artifact hash plus a range/recipe identity.
+- Derived sub-GGUF files need their own hash and a signed binding to the source artifact and range.
+- Rewriting a large GGUF for every placement is expensive and duplicates storage.
+- A range-aware mmap loader from one shared artifact remains preferable long term.
+
+Sub-GGUFs are still acceptable for a first integration spike because they already work.
+
+### 5.3 llama.cpp patch mechanics
+
+Nakshatra's patch series:
+
+- Adds range and endpoint fields to the model.
+- Reads namespaced partial-model metadata.
+- Allows unowned tensors to be absent.
+- Creates/loads only owned layer tensors.
+- Iterates only the selected layer interval.
+- Returns the unnormalized residual stream from non-tail stages.
+- Applies final norm and LM head only on the tail.
+
+Evidence:
+
+- `experiments/v0.0/m4_patches/llama-model.h.patch:1-17`.
+- `experiments/v0.0/m4_patches/llama-model.cpp.patch:1-73`.
+- `experiments/v0.0/m4_patches/llama-model-loader.cpp.patch:1-11`.
+- `experiments/v0.0/m4_patches/llama-graph.cpp.patch:1-24`.
+- `experiments/v0.0/m4_patches/models_llama.cpp.patch:1-70`.
+
+This is direct proof that the small-fork direction is feasible. The current patch is architecture-specific and built against older llama.cpp revisions. It should be rebased rather than copied blindly.
+
+### 5.4 Worker shape
+
+Each Python gRPC worker supervises one long-lived C++ daemon. The daemon owns the patched llama model/context and communicates with Python through a framed stdin/stdout or shared-memory protocol.
+
+Evidence:
+
+- `scripts/worker.py:1-17`.
+- `experiments/v0.0/worker_daemon.cpp:1-52`.
+- `experiments/v0.0/worker_daemon.cpp:166-267`.
+- `experiments/v0.0/worker_daemon.cpp:282-396`.
+
+The daemon accepts:
+
+- Token IDs for a head stage.
+- F32 boundary activations for middle/tail stages.
+- `start_pos` and `keep_kv` controls.
+
+It returns:
+
+- F32 residual activations for non-tail stages.
+- Greedy token IDs on the tail.
+- Optional all-position top tokens for speculative verification.
+
+The implementation also contains optional blockwise int8 activation transport:
+
+- `experiments/v0.0/worker_daemon.cpp:105-149`.
+
+This process boundary is close to the selected Meshnet worker shape and avoids exposing llama.cpp internal ABI to Python.
+
+### 5.5 Protobuf and route semantics
+
+Nakshatra's protobuf includes:
+
+- Protocol/backend/model/range capability reporting.
+- A source-model hash field.
+- Head/tail ownership flags.
+- Stable `session_id` and idempotency `step_id`.
+- Prefix/KV position metadata.
+- Token, hidden-state, logits, and error variants.
+- Server-to-server next-hop chains.
+- KV truncation.
+- Sleep/wake lifecycle operations.
+
+Evidence:
+
+- `proto/nakshatra.proto:1-55`.
+- `proto/nakshatra.proto:57-93`.
+- `proto/nakshatra.proto:95-131`.
+
+Meshnet's worker contract still needs additional fields:
+
+- Route epoch.
+- Effective overlap-safe start layer.
+- Exact source and derived artifact hashes.
+- Architecture/runtime/quantization/activation recipe.
+- Request/work/accounting identity.
+- Payload checksum and compression recipe.
+- Cache expectation and explicit cache-miss response.
+- Cancellation and lease identity.
+
+The implementation should sit behind a Meshnet-owned protocol rather than adopting the protobuf as a permanent public API.
+
+### 5.6 KV ownership and concurrency gap
+
+The daemon supports:
+
+- Cold prefill.
+- Incremental decode.
+- KV truncation after speculative rejection.
+- Sleep/wake and model reload.
+
+Evidence:
+
+- `experiments/v0.0/worker_daemon.cpp:344-384`.
+- `experiments/v0.0/worker_daemon.cpp:416-463`.
+- `experiments/v0.0/worker_daemon.cpp:479-500`.
+
+However, the audited worker creates one llama context with two sequence slots:
+
+```text
+sequence 0: serving/verification
+sequence 1: EAGLE scratch
+```
+
+The v0.5 design explicitly states that the daemon still has one logical serving session and ships with `n_concurrent_sessions = 1`:
+
+- `docs/v0.5-design-lock.md:97-111`.
+
+The Python idempotency cache deduplicates `(session_id,step_id)` outputs. It does not isolate independent llama KV state for multiple concurrent route sessions.
+
+Meshnet integration must add:
+
+```text
+(route_session_id, route_epoch)
+    -> llama_seq_id or isolated context
+```
+
+and verify concurrent prefill/decode, release, eviction, stale epoch rejection, and cache misses.
+
+### 5.7 Real acceptance evidence
+
+Nakshatra records a two-physical-machine CPU-only test over Tailscale:
+
+- Head worker: layers `[0,14)`.
+- Tail worker: layers `[14,28)`.
+- Prompt: “The capital of France is”.
+- Distributed first token: `12366`, “ Paris”.
+- Matched localhost, single-process chain, and whole-model llama.cpp reference.
+
+Evidence:
+
+- `experiments/v0.0/m6_findings.md:1-42`.
+
+This proves real independent layer execution and cross-machine activation transport. It does not prove GPU heterogeneity because both machines used CPU for the acceptance run.
+
+The repository also records a four-worker Llama-3.3-70B chain using Macs with Metal and one CPU stage:
+
+- Streaming completed around 0.21 token/s.
+- Multi-hop server push completed around 0.19 token/s.
+- Push was slower because compute dominated network time.
+- First generated token was “Paris”.
+- Alternate-worker replay completed with expected post-splice divergence.
+
+Evidence:
+
+- `docs/v0.5-design-lock.md:145-175`.
+- `docs/v0.5-design-lock.md:258-277`.
+
+The five-machine ROCm + Metal cross-vendor acceptance remained pending in the audited snapshot. Claims of fully validated ROCm+Metal execution should therefore remain qualified.
+
+### 5.8 Failure and numerical behavior
+
+Nakshatra supports:
+
+- Persistent streaming RPCs.
+- Server-to-server activation push.
+- Fallback from failed push to client relay.
+- Full-history replay after stream failure.
+- Alternate workers per range.
+- Drift-class-aware recovery design.
+
+Evidence:
+
+- `scripts/client.py:163-269`.
+- `scripts/client.py:842-850`.
+- `docs/v0.5-design-lock.md:258-277`.
+- `docs/v1.0-fault-tolerance.md:93-103`.
+
+The project correctly acknowledges that replay on another backend can numerically diverge. For Meshnet, the safe default remains:
+
+1. Exact recipe and same drift class for in-session replacement.
+2. Otherwise restart from token zero on a newly consistent route.
+3. Never silently import incompatible KV or continue with an unvalidated mixed recipe.
+
+### 5.9 Security and work receipts
+
+The worker contains optional:
+
+- TLS and SPKI pinning.
+- Ed25519 request authentication.
+- Admission and sandbox hooks.
+- Audit logging.
+- An experimental encrypted fabric.
+
+Evidence:
+
+- `scripts/worker.py:49-117`.
+- `scripts/worker.py:218-319`.
+
+The receipt implementation explicitly documents that:
+
+- Output hash and structural consistency are independently checkable.
+- Model identity is only asserted because the live model hash is a zero stub.
+- Participation is coordinator-asserted because worker signatures are not populated.
+
+Evidence:
+
+- `scripts/receipt.py:1-27`.
+- `scripts/receipt.py:34-47`.
+- `scripts/receipt.py:50-97`.
+- `scripts/receipt.py:100-160`.
+
+This is not sufficient for per-node rewards. Meshnet must bind receipts to:
+
+```text
+request/work ID
+route session and epoch
+source artifact hash
+layer range and effective start
+input/output activation digests
+positions/token count
+runtime recipe
+measured compute
+worker identity/signature
+```
+
+### 5.10 Reproducibility and integration defects
+
+The audited repository is not a self-contained llama.cpp fork or reproducible worker build:
+
+- It ships patch files but no pinned llama.cpp source tree or submodule.
+- The README recommends llama.cpp commit `c46583b` “or close enough”.
+- The daemon is copied manually into an external llama.cpp checkout and its CMake target is added manually.
+- The documented copy recipe omits `shm_ring.hpp`, which the daemon includes.
+- The historical two-machine run used different llama.cpp builds on the two hosts.
+
+Evidence:
+
+- `README.md:41-77`.
+- `experiments/v0.0/worker_daemon.cpp:54-57`.
+- `experiments/v0.0/m6_findings.md:35-42`.
+
+The live daemon also trusts operator-supplied range and endpoint mode rather than returning authoritative metadata from the loaded sub-GGUF. Its INFO response reports a generic full range and both endpoint flags, while Python advertises CLI values:
+
+- `experiments/v0.0/worker_daemon.cpp:387-392`.
+- `experiments/v0.0/worker_daemon.cpp:466-475`.
+- `scripts/worker.py:1141-1153`.
+
+Additional integration gaps:
+
+- Normal activations are little-endian F32 even though the protobuf comment says FP16 is the default; int8 activation mode is environment-global rather than negotiated.
+- The coordinator requires full-GGUF access through `llama-cpp-python` for tokenization.
+- Tail sampling is hard-coded greedy top-1 rather than returning general logits.
+- Many tests use fake daemons; no CI job builds patched llama.cpp, generates slices, starts two real workers, and tests prefill/decode/failure.
+- Nakshatra runtime dependencies are absent from the inherited Petals package metadata.
+
+Evidence:
+
+- `proto/nakshatra.proto:11-12`.
+- `scripts/worker.py:340-343`.
+- `scripts/worker.py:1141-1151`.
+- `scripts/client.py:129-139`.
+- `scripts/client.py:719-722`.
+- `experiments/v0.0/worker_daemon.cpp:617-641`.
+- `tests/test_worker_eagle_sleep_rpc.py:1-31`.
+- `setup.cfg:1-16`.
+- `setup.cfg:29-72`.
+
+These defects make Nakshatra a strong source donor and independent feasibility proof, but not the repository to fork as the production base.
+
+### 5.11 Reuse decision
+
+Do not recreate Nakshatra's working Llama-family layer patch and daemon from scratch without first attempting upstream collaboration.
+
+Preferred plan:
+
+1. Reproduce its two-worker path locally.
+2. Rebase its patch against the exact current llama.cpp commit already audited.
+3. Compare the patch with the proposed `llama_model_load_range` and boundary-output design.
+4. Borrow or jointly maintain narrow patch concepts and tests, but keep a project-owned standalone worker and small pinned llama.cpp fork rather than forking the Petals-derived repository wholesale.
+5. Replace or adapt its Python control plane with existing Meshnet tracker/session/relay/billing infrastructure.
+6. Add multi-session KV isolation, exact recipe identity, route epochs, cancellation, signed work receipts, and architecture certification.
+7. Upstream generic llama.cpp range-loading/boundary hooks where maintainers will accept them.
+
+Classification:
+
+```text
+Primary source donor and collaboration candidate
+Do not adopt or fork the repository wholesale
+```
+
+## 6. LiGGUF SARA
+
+### 6.1 Verdict
+
+LiGGUF contains a real experimental networked distributed implementation. It is not layer pipeline parallelism.
+
+Its SARA implementation performs tensor-parallel activation reduction across every transformer layer.
+
+### 6.2 Mechanism
+
+Every process mmaps and parses the complete GGUF and constructs pointers for all transformer blocks:
+
+- `cpp/ligguf_distrib.cpp:323-485`.
+
+Rank assignments split:
+
+- KV heads.
+- Query heads derived from KV heads.
+- FFN channel blocks.
+
+Evidence:
+
+- `cpp/ligguf_distrib.cpp:120-124`.
+- `cpp/ligguf_distrib.cpp:657-679`.
+
+Each rank allocates KV for its attention-head slice across every layer:
+
+- `cpp/ligguf_distrib.cpp:681-707`.
+
+For every transformer layer, the master:
+
+1. Normalizes the residual.
+2. Broadcasts a Q8_0 activation to all workers.
+3. Every rank computes its attention partial.
+4. Workers return Q8_0 full-width partials.
+5. The master sums partials into the residual.
+6. Repeats the same process for the FFN.
+
+Evidence:
+
+- `cpp/ligguf_distrib.cpp:709-767`.
+- `cpp/ligguf_distrib.cpp:892-903`.
+- `cpp/ligguf_distrib.cpp:957-974`.
+- `cpp/ligguf_distrib.cpp:1025-1066`.
+
+### 6.3 Strengths
+
+- Very compact and readable.
+- Direct GGUF mmap.
+- Real raw-TCP master/worker implementation.
+- Q8_0 activation and partial transport.
+- Local KV per attention-head shard.
+- Useful reference for tensor-parallel reductions on CPU-heavy edge systems.
+
+### 6.4 Mismatch
+
+- Every worker needs the complete GGUF.
+- Every worker participates in every layer.
+- Two network reduction phases occur per transformer layer.
+- Static rank/world topology.
+- One master connection and one generation at a time.
+- Sequential worker receive loop.
+- No model/session/request/range identity.
+- No authentication, checksums, recovery, cancellation, or accounting.
+- Custom inference core has much narrower model/kernel coverage than llama.cpp.
+
+Evidence:
+
+- `cpp/ligguf_distrib.cpp:769-974`.
+- `cpp/ligguf_distrib.cpp:1025-1046`.
+- `cpp/ligguf_distrib.cpp:1138-1241`.
+
+Decision: retain as a source donor for compact Q8 activation transport, tensor-partition tests, and reduction benchmarking. Do not use it as the native Meshnet layer worker.
+
+## 7. Chameleon
+
+Audited snapshot:
+
+```text
+megeezy/Chameleon
+commit 96fbd96a9f67d29d12292d3373c88996aba65f84
+```
+
+Chameleon is a whole-model lifecycle and routing system:
+
+- Coordinator selects a model and worker.
+- Python worker loads a complete llama-cpp-python, vLLM, Transformers, or ExLlamaV2 backend.
+- The model executes, can remain warm, and is later unloaded.
+
+It does not split one model's layers or tensors across workers. Its README also labels it design phase.
+
+Decision: exclude from the distributed-GGUF implementation list. Whole-model load/unload and warm-cache ideas may be relevant to proxy backends only.
+
+## 8. Continuum
+
+Audited snapshot:
+
+```text
+CambrianTech/continuum
+commit dd976df36079d75244719a23956e1c9e2dcddc27
+```
+
+Continuum has:
+
+- A local GGUF loader.
+- A local inference backend.
+- Mesh/federation infrastructure.
+- Roadmap language about dividing models across nodes.
+
+No executable distributed GGUF layer or tensor data plane was found in this snapshot. There is no layer-range protocol, boundary-activation transport, distributed KV ownership, or distributed GGUF test corresponding to the roadmap statement.
+
+Decision: exclude until source and real execution evidence exist.
+
+## 9. GitHub search conclusion
+
+The expanded searches used terms including:
+
+- `distributed GGUF`.
+- `sub-GGUF`.
+- `layer range` with llama.cpp.
+- `result_partial_hidden`.
+- `activation chain`.
+- llama.cpp RPC orchestration.
+- GGUF tensor parallelism.
+
+The working implementation families found are:
+
+| Family | Projects | What is real |
+|---|---|---|
+| Coordinator-owned remote devices | llama.cpp RPC, historical GPUStack/llama-box, LocalAI integrations | Full GGML graph controlled by one coordinator |
+| Independent GGUF layer pipeline | Nakshatra, prima.cpp | Local layer ownership, residual boundaries, local KV |
+| Custom Rust layer pipeline | `llama-gguf` | gRPC layers, but coordinator-streamed weights and weak session semantics |
+| Tensor parallel GGUF | LiGGUF SARA | Head/FFN partitions and per-layer partial reductions |
+| Static/custom non-GGUF tensor engines | distributed-llama/dllama and similar | Useful algorithms, different artifacts/runtime |
+| Whole-model routing | Chameleon, Ollama, most LocalAI use, current GPUStack backends | Independent complete models, not one split model |
+| Roadmap-only mesh claims | Continuum and several search hits | No executable distributed GGUF data plane found |
+
+No additional mature project was found that already combines:
+
+- Arbitrary tracker-selected layer intervals.
+- Exact local GGUF artifact ownership.
+- Heterogeneous volunteer nodes.
+- Concurrent route-session KV.
+- Dynamic route epochs and recovery.
+- Relay and cancellation.
+- Exact runtime/activation compatibility.
+- Worker-authenticated accounting.
+
+Nakshatra is the closest and materially reduces the amount of new inference-engine work required.
+
+## 10. Revised architecture decision
+
+The selected runtime remains llama.cpp/GGML, but the implementation source priority changes:
+
+```text
+Existing Meshnet tracker, relay, sessions, capability admission, and billing
+                              |
+                Meshnet-owned stable shard protocol
+                              |
+       project-owned C++ worker borrowing narrow Nakshatra concepts/tests
+                              |
+       narrow, pinned, architecture-certified llama.cpp patch
+                              |
+       GGUF loader, quant kernels, KV, CPU/GPU backends
+```
+
+This is not an architectural pivot away from the initial decision. Nakshatra is an external implementation of nearly the same missing seam.
+
+## 11. Revised implementation sequence
+
+### 11.1 Reproduction and patch comparison
+
+- Reproduce Nakshatra's two local workers with a small dense Llama-family GGUF.
+- Verify disjoint sub-GGUF weight ownership.
+- Verify boundary residual parity with whole-model llama.cpp.
+- Rebase against the pinned current llama.cpp commit.
+- Compare pre-sliced GGUF loading with range-aware loading from one source artifact.
+
+### 11.2 Meshnet protocol adapter
+
+- Keep the project-owned llama.cpp worker behind a supervised executable.
+- Translate Meshnet request/session/route metadata into worker calls.
+- Preserve BF16 or versioned named-tensor bundles on the public network boundary.
+- Add exact source/slice/runtime recipe checks.
+- Reject stale route epochs and incompatible caches.
+
+### 11.3 Multi-session KV
+
+- Map each route session/epoch to an isolated `llama_seq_id` or context.
+- Test concurrent prefill and decode.
+- Add bounded release, TTL, LRU, and cache-miss semantics.
+- Verify only owned layers allocate KV.
+
+### 11.4 Real heterogeneous route
+
+- CPU plus AMD HIP/ROCm first on the available machine.
+- Add CUDA, Vulkan, and Metal as certified lanes when hardware is available.
+- Measure numerical drift and define compatibility classes.
+- Require a clean from-token-zero restart when exact-recipe recovery is unavailable.
+
+### 11.5 Trustworthy accounting
+
+- Worker signs work receipts.
+- Receipt binds route, request, artifact, range, activation digests, positions, and runtime recipe.
+- Tracker validates structural consistency and completion evidence before rewards.
+
+### 11.6 Architecture expansion
+
+- Dense Llama-family first.
+- Explicit Qwen3/Qwen3-MoE adapter next.
+- Fail closed for all unvalidated architectures.
+
+## 12. Local validation performed
+
+The source audit included limited executable verification without downloading model artifacts:
+
+```text
+LiGGUF distributed target:
+make ligguf-cpp-distrib
+PASS — g++ produced ligguf-cpp-distrib
+
+Nakshatra dependency-free focused tests:
+47 passed in 0.42s
+```
+
+The Nakshatra subset covered receipt validation, wire-version behavior, wire handshake behavior, and topology ordering.
+
+The networked idempotency integration test was not collected because this audit environment does not have the optional `grpcio` package installed. This is an environment dependency blocker, not a test assertion failure. No real Nakshatra model inference was run locally during this research; the real-model conclusions remain tied to the source, recorded experiment evidence, and the future reproduction gate in section 11.
+
+## 13. Final conclusions
+
+1. The user's central practical observation is correct: llama.cpp is the base of most real-world GGUF distribution found.
+2. GPUStack 0.4 was open source and genuinely distributed, but through llama-box/llama.cpp RPC rather than independent shards.
+3. GPUStack 2.0 removed distributed GGUF support, so the old tutorial must be treated as historical.
+4. Nakshatra is the most important new source. It has already implemented and exercised the narrow llama.cpp layer-worker design.
+5. Nakshatra should be approached for narrow collaboration and mined for source/tests, but its repository should not be adopted or forked wholesale.
+6. Nakshatra still needs Meshnet-specific hardening: exact identity, concurrent KV, route epochs, cancellation, accounting, and architecture certification.
+7. LiGGUF SARA is real distributed GGUF tensor parallelism, but every worker loads the whole model and communicates twice per layer.
+8. Chameleon is whole-model routing; Continuum's distributed GGUF is roadmap-only in the audited snapshot.
+9. No drop-in project yet satisfies the complete tracker-routed volunteer-network contract.
+10. The implementation risk is now lower because the hardest first proof—partial llama.cpp layer loading plus boundary execution—has independent working source and cross-machine evidence.

docs/research/distributed-gguf-landscape.md

@@ -0,0 +1,829 @@
+# Distributed GGUF inference: existing projects, source audits, and implementation direction
+
+Status: Research complete; architecture direction selected
+Last updated: 2026-07-13
+
+## 1. Purpose
+
+This document records the research behind native distributed GGUF inference for the neuron-tai network. It is intentionally not a design for proxying an already-running whole-model `llama-server`, Ollama, vLLM, or another OpenAI-compatible runtime. Whole-model proxying is useful as a correctness/performance baseline, but it does not solve the network's central problem.
+
+The required product is a tracker-routed distributed GGUF data plane in which heterogeneous nodes independently execute model layer ranges, exchange boundary activations, retain cache state for their own layers, and receive credit for work they actually perform.
+
+### 1.1 Required contract
+
+A satisfactory implementation must support:
+
+- Native GGUF model artifacts and llama.cpp-compatible quantizations.
+- Independently loaded, executable layer ranges on each node.
+- Tracker-selected routes rather than one static rank topology.
+- Heterogeneous CPU, CUDA, HIP/ROCm, Vulkan, Metal, and other certified lanes.
+- A versioned hidden-state activation boundary between nodes.
+- Local per-shard KV or recurrent state keyed by route session.
+- Prefill and decode phases with bounded wire and compute cost.
+- Overlap-safe execution using the tracker's effective `start_layer` semantics.
+- Explicit cache miss, eviction, route epoch, and recovery behavior.
+- Dynamic health and route failure handling.
+- Per-node work telemetry and accounting.
+- Model-agnostic infrastructure with architecture-specific certification.
+
+### 1.2 Non-goals
+
+- Reimplementing GGUF parsing, quantization kernels, tokenization, or mature CPU/GPU kernels from scratch.
+- Treating a whole-model node as the distributed solution.
+- Treating stock llama.cpp RPC as a safe volunteer-network protocol.
+- Claiming every GGUF architecture works without a bounded real distributed validation.
+- Mixing Transformers and GGUF shards in one route without an explicit compatible activation contract.
+
+### 1.3 Parallelism taxonomy
+
+These mechanisms must not be conflated:
+
+- **Layer or pipeline parallelism** assigns whole contiguous transformer-layer ranges to different stages and transports boundary activations between them. This is the closest match to neuron-tai's tracker-routed shard contract.
+- **Tensor parallelism** partitions operations or tensors within each layer, such as attention heads, matrix rows/columns, or experts. It usually requires all ranks for every layer and collective or reduction traffic inside every layer.
+- **Local multi-device offload** places tensors or layers across devices controlled by one process. llama.cpp's `n_gpu_layers` and `tensor_split` belong here unless RPC devices are included.
+- **llama.cpp RPC offload** exposes remote GGML devices to a coordinator-owned graph. It is cross-machine, but remote processes are devices rather than independent model/session workers.
+- **Whole-model replication** runs one complete model per server and load-balances requests. It increases throughput and availability but does not allow one oversized model to span workers.
+
+GGUF weight quantization such as Q4_K_M or Q8_0 reduces storage and memory pressure. It does not define the activation dtype, compute dtype, KV-cache dtype, or distributed topology.
+
+## 2. Existing neuron-tai infrastructure
+
+The repository already contains most of the control plane needed by a GGUF shard worker.
+
+### 2.1 Backend and execution path
+
+There is no formal backend interface yet. Production code uses the concrete `TorchModelShard`, while several call sites rely on duck typing.
+
+Relevant source:
+
+- `packages/node/meshnet_node/model_backend.py:72-223` — result types, session cache, and concrete backend.
+- `packages/node/meshnet_node/model_backend.py:226-345` — Torch/Hugging Face construction.
+- `packages/node/meshnet_node/model_backend.py:347-566` — effective head, middle, tail, and whole-model backend methods.
+- `packages/node/meshnet_node/model_backend.py:730-747` — Torch factory.
+- `packages/node/meshnet_node/torch_server.py:302-317` — injected backend usage.
+- `packages/node/meshnet_node/torch_server.py:717-766` — whole-model generation fast path.
+- `packages/node/meshnet_node/torch_server.py:1464-1514` — server construction.
+- `packages/node/meshnet_node/torch_server.py:1638-1659` — factory hard-wired to `load_torch_shard`.
+
+The current effective distributed backend contract includes:
+
+- `model_id`, `shard_start`, `shard_end`, `total_layers`, `is_head`, `is_tail`, and `device`.
+- `encode_prompt`, `encode_next_token`, `forward_bytes`, and `decode_tail_token`.
+- `eos_token_ids` and `release_session`.
+- Whole-model generation and token-count helpers.
+
+A GGUF design should separate a generic backend/process lifecycle interface from an optional activation-shard interface rather than making every backend pretend to be `TorchModelShard`.
+
+### 2.2 Distributed activation wire
+
+The current data plane already provides a reusable envelope:
+
+- Binary `POST /forward` requests.
+- `X-Meshnet-Session` route-session identity.
+- `X-Meshnet-Cache: prefill | decode`.
+- `X-Meshnet-Past-Len` cache consistency check.
+- `X-Meshnet-Start-Layer` overlap-safe execution.
+- Explicit tensor shape, dtype, position IDs, attention mask, chunk, and compression metadata.
+- Direct and relay downstream clients owned by the generation handler.
+- HTTP 409 cache-miss responses and re-prefill recovery.
+
+Relevant source:
+
+- `packages/node/meshnet_node/server.py:13-76` — wire contract and validation.
+- `packages/node/meshnet_node/torch_server.py:497-635` — binary handler.
+- `packages/node/meshnet_node/torch_server.py:974-1288` — route parsing and hop execution.
+- `packages/tracker/meshnet_tracker/server.py:3635-3781` — route planning and effective start layers.
+- `docs/adr/0008-binary-activation-wire-format.md`.
+- `docs/adr/0012-start-layer-overlapping-shards.md`.
+
+The current payload semantics are still Torch/Hugging Face-specific:
+
+- Boundary dtype is fixed to BF16.
+- Tensor conversion imports Torch.
+- Attention masks and position IDs use Torch-oriented serialization.
+- Tail-local decode reconstructs a Torch tensor.
+
+The HTTP envelope is reusable; backend-neutral activation semantics need to be made explicit.
+
+### 2.3 Local shard cache
+
+The existing Transformers path already has the desired session behavior:
+
+- One stable UUID per distributed generation.
+- Prefill establishes state on every shard.
+- Decode forwards only the new-token activation.
+- Every shard stores only its own layer state.
+- Cache lookup checks sequence length and effective start layer.
+- Cache miss, restart, or route mismatch returns HTTP 409.
+- The head re-prefills accumulated tokens after a miss.
+- TTL and LRU bound local memory.
+
+Relevant source:
+
+- `packages/node/meshnet_node/model_backend.py:102-193`.
+- `packages/node/meshnet_node/model_backend.py:334-345`.
+- `packages/node/meshnet_node/model_backend.py:595-662`.
+- `packages/node/meshnet_node/torch_server.py:806-944`.
+- `docs/adr/0022-sharded-per-node-kv-cache.md`.
+
+A GGUF worker should preserve this product-level contract while mapping route sessions to llama.cpp sequence IDs or isolated contexts.
+
+### 2.4 Tracker and capability admission
+
+The tracker already supports:
+
+- Model and layer-range registration.
+- Capability reports and fail-closed admission.
+- Heterogeneous route candidates.
+- Greedy interval coverage.
+- Effective start-layer injection.
+- Throughput/load/reputation-informed selection.
+- Dynamic health and route statistics.
+- Accounting and work attribution.
+
+Relevant source:
+
+- `packages/tracker/meshnet_tracker/server.py:592-942`.
+- `packages/tracker/meshnet_tracker/server.py:3635-3781`.
+- `packages/tracker/meshnet_tracker/server.py:4425-4595`.
+- `packages/tracker/meshnet_tracker/server.py:6200-6285`.
+- `docs/adr/0021-dynamic-statistical-routing.md`.
+- `docs/adr/0023-model-agnostic-node-capability-admission.md`.
+
+GGUF gaps include:
+
+- The closed `bfloat16`, `int8`, `nf4` precision vocabulary.
+- No first-class route kind or activation compatibility key.
+- No GGUF artifact descriptor.
+- No llama.cpp/Vulkan/HIP execution recipe.
+- No model-aware GGUF memory estimator.
+- No real llama.cpp doctor benchmark.
+
+## 3. Ecosystem survey
+
+No mature project was found that combines native GGUF, arbitrary tracker-selected layer ranges, heterogeneous cross-machine execution, independent local shard cache, dynamic route sessions, recovery, and per-node accounting.
+
+A later GitHub follow-up found that [Nakshatra](https://github.com/fthrvi/nakshatra) already implements the narrow patched-llama.cpp Llama-family layer-worker seam and should be treated as the primary source donor and collaboration candidate. Its repository is not self-contained or production-ready enough to adopt as the runtime base. Historical GPUStack/llama-box and LiGGUF SARA were also audited in [the GitHub follow-up](distributed-gguf-github-followup.md). This reduces implementation risk but does not provide a drop-in Meshnet runtime.
+
+### 3.1 Candidate summary
+
+| Project | Native GGUF | Distributed form | Direct fit | Decision |
+|---|---:|---|---|---|
+| [llama.cpp](https://github.com/ggml-org/llama.cpp) | Yes | Device offload/RPC, local multi-GPU | Best kernel, wrong stock topology | Use as primary kernel through a small pinned fork |
+| [Nakshatra](https://github.com/fthrvi/nakshatra) | Yes, via sub-GGUFs and patched llama.cpp | Independent contiguous layer workers over gRPC | Closest implementation found | Primary source donor/collaboration candidate; do not adopt its repository wholesale |
+| [llama-gguf](https://github.com/Lexmata/llama-gguf) | Yes | gRPC layer pipeline | Architecturally close, immature custom runtime | Source donor only |
+| [prima.cpp](https://github.com/OpenCPIL/prima.cpp) | Yes | Static piped-ring layer windows | Proves partial loading and local KV | Source donor only |
+| [LiGGUF](https://github.com/matrixsmaster/ligguf) | Yes | SARA head/FFN tensor sharding and activation reduction | Every rank loads the full model; static master/workers | Compact tensor-parallel source donor only |
+| [mistral.rs](https://github.com/EricLBuehler/mistral.rs) | Yes | NCCL TP and multi-machine ring | Static/homogeneous distributed assumptions | Evaluate as optional homogeneous backend |
+| [vLLM](https://github.com/vllm-project/vllm) | Experimental/plugin | TP, PP, DP, EP via Ray/Torch distributed | Production clusters, not volunteer layer routes | Whole-model/managed-cluster lane and design donor; see [vLLM assessment](vllm-distributed-gguf-assessment.md) |
+| [distributed-llama](https://github.com/b4rtaz/distributed-llama) | No, custom format | Root/worker tensor parallelism | Power-of-two/static topology | Ideas only |
+| [Petals](https://github.com/bigscience-workshop/petals) | No | Internet layer pipeline | Strong cache/session semantics | Conceptual donor |
+| [exo](https://github.com/exo-explore/exo) | No native path | MLX tensor/pipeline sharding | MLX-centric and platform-asymmetric | Placement/cache ideas only |
+| [LocalAI](https://github.com/mudler/LocalAI) | Via llama.cpp | RPC workers and request routing | Control-plane overlap | Lifecycle ideas only |
+| [GPUStack](https://github.com/gpustack/gpustack) 0.4-0.7 | Via llama-box/RPC | Scheduler + primary/RPC workers | Proven RPC orchestration, not layer workers | Historical reference; GGUF distribution removed in GPUStack 2.0 |
+| [llama-box](https://github.com/gpustack/llama-box) | Yes | llama.cpp RPC | Archived | Historical source donor only |
+| [Chameleon](https://github.com/megeezy/Chameleon) | Via whole-model backends | Whole-model worker lifecycle/routing | No single-model sharding | Exclude from distributed-GGUF candidates |
+| [Continuum](https://github.com/CambrianTech/continuum) | Local loader/backend | Mesh orchestration; tensor distribution is roadmap text | No distributed GGUF execution path found | Exclude until executable evidence exists |
+| Ollama | Yes | Local multi-device; no cross-host model split | Whole-model only | Proxy baseline only |
+| MLC LLM | No native GGUF | Compiled local/multi-GPU | Different artifact/runtime | Separate recipe at most |
+
+### 3.2 llama.cpp RPC
+
+Stock RPC exposes remote GGML devices to one coordinator. Current upstream documentation says it can distribute weights and KV across local and remote devices and can cache remote tensors with `ggml-rpc-server -c`. This makes the original claim in `docs/adr/0001-pytorch-over-llama-cpp.md`—that every launch must always resend all weights—outdated.
+
+RPC still does not provide:
+
+- Independently loaded local GGUF layer workers.
+- Tracker-selected per-request routes.
+- A hidden-state HTTP endpoint.
+- Route-session-owned shard cache.
+- Per-worker tracker accounting.
+- A safe untrusted volunteer-node boundary.
+
+Upstream explicitly calls RPC proof-of-concept, fragile, and insecure. It is not the target architecture.
+
+### 3.3 Petals
+
+Petals remains the strongest conceptual reference for:
+
+- Hosting independent transformer block ranges.
+- Per-server cache state.
+- Session-based inference.
+- Rebuilding cache after route failure.
+- Capacity-aware block placement.
+- Public/private swarm control.
+
+It is PyTorch/Transformers-based, not GGUF, and is no longer active enough to adopt as the runtime. The existing neuron-tai Transformers path already implements many of its core semantics.
+
+## 4. Source audit: prima.cpp
+
+Audited source snapshot:
+
+```text
+OpenCPIL/prima.cpp
+commit 6f9b7c40962d777d1726456b4359340d932bef12
+```
+
+### 4.1 Decision
+
+Do not fork prima.cpp wholesale. Extract partial-loading, local-KV, graph-boundary, and placement ideas into a small fork of current llama.cpp.
+
+prima.cpp is a full invasive llama.cpp fork. Its exact upstream ancestry could not be established from the available repository history. Distributed changes are embedded in public structures, `src/llama.cpp`, common CLI code, and the old server.
+
+### 4.2 Layer ownership
+
+prima.cpp adds static rank topology fields:
+
+- `n_world`.
+- `rank`.
+- Fixed `n_layer_window[32]` arrays.
+- Cycle count and fixed network addresses/ports.
+
+Evidence:
+
+- `include/llama.h:292-353`.
+- `src/llama.cpp:2597-2636`.
+- `common/arg.cpp:680-785`.
+
+Layer ownership is a repeated cyclic window, not an arbitrary tracker-provided `[start,end)` route:
+
+- `src/llama.cpp:3838-3883`.
+- `src/llama.cpp:16941-16951`.
+
+This cannot directly represent dynamic tracker routes.
+
+### 4.3 Partial local GGUF loading
+
+This is prima.cpp's strongest reusable contribution.
+
+- Rank zero creates token embedding, final norm, and output tensors.
+- Transformer tensors are created only when the layer belongs to the process.
+- Global layer IDs are compacted into local storage.
+- Unneeded GGUF tensors are erased before mapping/loading.
+- Only selected file regions are mapped and loaded.
+
+Evidence:
+
+- `src/llama.cpp:7500-7619`.
+- `src/llama.cpp:9358-9515`.
+
+This proves a llama.cpp-derived process can independently load only the GGUF tensors required by its layer range.
+
+### 4.4 Graph boundaries
+
+prima.cpp introduces explicit intermediate input/output tensors and skips unowned layers:
+
+- `src/llama.cpp:10793-10813`.
+- `src/llama.cpp:11000-11117`.
+- `src/llama.cpp:16953-17010`.
+
+The distributed graph path is hard-asserted to Llama or Qwen2 despite inherited architecture enums. Architecture support is therefore narrow.
+
+### 4.5 Activation transport
+
+Transport is raw ZeroMQ multipart PUSH/PULL. Messages include native shapes and raw contiguous F32 activation bytes.
+
+Evidence:
+
+- `src/llama.cpp:18031-18077`.
+- `src/llama.cpp:18542-18563`.
+
+It has no protocol version, model/session/route identity, layer range, dtype field, checksum, authentication, compression, or robust payload validation. It is incompatible with the existing Meshnet BF16 route-session protocol.
+
+### 4.6 Local KV
+
+Each process has one local llama.cpp context and allocates KV only for layers assigned to that process.
+
+Evidence:
+
+- `src/llama.cpp:3889-3969`.
+- `src/llama.cpp:18268-18269`.
+
+This is valuable physical behavior, but cache operations are coupled to one static ring and server slot numbering. There is no route session, route epoch, model fingerprint, range identity, lease, or idempotency contract.
+
+### 4.7 Fault model and backends
+
+Runtime sends and receives are blocking. Socket failures can terminate or stall a process. There is no runtime route repair, KV replay, deduplication, or accounting.
+
+The inherited tree contains many backends, but the distributed profiler and placement solver primarily model CPU, CUDA, and Metal. Project documentation excludes AMD/Vulkan distributed support.
+
+Evidence:
+
+- `src/llama.cpp:18031-18077`.
+- `src/llama.cpp:20492-20769`.
+- `common/profiler.h:329-405`.
+- `common/common.cpp:850-1183`.
+- `README.md:362-368`.
+
+### 4.8 Tests
+
+No meaningful distributed regression suite was found for layer windows, transport, cache propagation, heterogeneous placement, session isolation, or failure behavior.
+
+### 4.9 Reusable parts
+
+Port or use as design references:
+
+- Selective endpoint/head/tail tensor creation.
+- Selected-layer GGUF mapping and loading.
+- Global-to-local layer indexing.
+- Per-layer KV filtering and allocation.
+- Boundary residual input/output.
+- Placement cost ideas and page prefetching.
+
+Do not reuse:
+
+- Static ring topology.
+- Fixed rank arrays.
+- ZeroMQ wire protocol.
+- Server patches.
+- Cache-command propagation.
+- Failure semantics.
+
+## 5. Source audit: llama-gguf
+
+Audited source snapshot:
+
+```text
+Lexmata/llama-gguf
+commit 6e9f194206450080d47101c6f88a80f604ce69be
+```
+
+### 5.1 Decision
+
+Do not adopt `llama-gguf` as the inference runtime or distributed data plane. Use it as a source donor for protobuf organization, health/capability messages, explicit ranges, and synthetic multi-process tests.
+
+### 5.2 Coordinator owns and streams the model
+
+The coordinator loads and builds the entire GGUF model, then serializes layer tensors and streams them to shards over gRPC.
+
+Evidence:
+
+- `src/distributed/coordinator.rs:44-53`.
+- `src/distributed/coordinator.rs:160-168`.
+- `src/distributed/coordinator.rs:196-299`.
+
+Workers explicitly do not open local GGUF files:
+
+- `src/distributed/shard.rs:53-59`.
+
+This conflicts with independently owned local artifacts, startup efficiency, and tracker-controlled model distribution.
+
+### 5.3 Explicit ranges but simple coverage validation
+
+The cluster config supports explicit half-open ranges or even auto-partitioning:
+
+- `src/distributed/config.rs:65-92`.
+- `src/distributed/config.rs:168-220`.
+
+Manual validation checks total assigned layer count but does not prove sorted, gap-free, non-overlapping coverage. It is a static startup configuration, not a dynamic per-request route.
+
+### 5.4 Shard-local KV, but only one global sequence
+
+A worker allocates one KV cache for its local layer count:
+
+- `src/distributed/shard.rs:30-44`.
+- `src/distributed/shard.rs:261-286`.
+- `src/model/mod.rs:62-117`.
+
+There is no cache map or route-session identity. `ResetKvCache` takes an empty request and resets the one global cache:
+
+- `src/distributed/shard.rs:447-458`.
+- `proto/distributed.proto:15-16`.
+- `proto/distributed.proto:95-103`.
+
+Concurrent sessions, cache epochs, stale requests, and per-request eviction are unsupported.
+
+### 5.5 Forward protocol
+
+`ForwardRequest` contains only hidden state, position, and sequence length:
+
+- `proto/distributed.proto:77-93`.
+
+It has no model fingerprint, route/session/request identity, layer range, phase, chunk, expected cache length, accounting identity, idempotency key, or route epoch.
+
+The tensor envelope does include shape, dtype, little-endian bytes, and a name:
+
+- `proto/distributed.proto:36-46`.
+- `src/distributed/tensor_transfer.rs:10-115`.
+
+The DType enum can represent BF16, but the distributed execution path and tests use F32 hidden states and F32 KV.
+
+### 5.6 Token-by-token serialized execution
+
+The distributed model loops over tokens one at a time and sends each hidden vector through each shard. The complete pipeline is protected by one mutex.
+
+Evidence:
+
+- `src/distributed/model.rs:21-39`.
+- `src/distributed/model.rs:86-147`.
+- `src/distributed/pipeline.rs:45-113`.
+
+This prevents efficient batched prefill and safe session multiplexing.
+
+### 5.7 Architecture claims versus shard reconstruction
+
+The repository has a large architecture enum, but the distributed shard reconstructs a simple dense layer consisting of RMSNorm, ordinary attention, and dense gated FFN.
+
+Evidence:
+
+- `src/model/architecture.rs:5-155`.
+- `src/distributed/shard.rs:154-223`.
+
+The worker cannot reconstruct many architecture-specific layer forms such as MoE, hybrid/recurrent layers, shared cache, architecture-specific norms, or specialized projections. Qwen3 MoE support cannot be inferred from the enum list.
+
+### 5.8 Backends
+
+Shard backend selection tries CUDA, Metal, DX12, Vulkan, then CPU, but the source notes GPU model-weight preloading is skipped because weights arrive as gRPC tensors. The per-operation backend path is not equivalent to llama.cpp's mature graph scheduling.
+
+Evidence:
+
+- `src/distributed/shard.rs:53-115`.
+- `src/backend/mod.rs:25-260`.
+
+### 5.9 Fault recovery
+
+The repository has health monitoring and can reconnect, reconfigure, and resend layers:
+
+- `src/distributed/fault.rs:83-360`.
+
+It does not restore active KV, route position, partial prefill, session ownership, or accounting state. Reloading a process is not active-generation recovery.
+
+### 5.10 Tests
+
+Distributed integration tests start local CPU gRPC shards with synthetic dense F32 weights. They verify configure, load, forward, reset, and a two-shard pipeline.
+
+The file explicitly states that no real GGUF is loaded:
+
+- `tests/distributed_integration_test.rs:1-7`.
+
+Missing evidence includes real quantized GGUF parity, heterogeneous devices, BF16 boundaries, batched prefill, concurrent sessions, worker loss, route epochs, local artifact loading, and cache recovery.
+
+### 5.11 Reusable parts
+
+Potential donors:
+
+- Protobuf tensor shape/dtype/data envelope.
+- Explicit half-open layer ranges.
+- Health and capabilities messages.
+- Synthetic multi-process shard test harness.
+- Separation of configure, load, forward, reset, and health methods.
+
+Do not reuse:
+
+- Coordinator-side weight streaming.
+- Custom inference engine as the production kernel.
+- Single global KV cache.
+- Token-by-token pipeline.
+- Dense-only layer reconstruction.
+- Fault-reload semantics.
+- The all-reduce implementation, which currently echoes its input at `src/distributed/shard.rs:522-541`.
+
+## 6. Source audit: current llama.cpp
+
+Audited source snapshot:
+
+```text
+ggml-org/llama.cpp
+commit 91c631b21d6e5d09e9c6659efdf6baeef5a44ddb
+```
+
+### 6.1 Decision
+
+The smallest credible implementation is a pinned, architecture-certified layer-worker fork of current llama.cpp. A fully generic all-architecture patch is not credible as the first implementation.
+
+### 6.2 Layer-range API
+
+Prefer a new range loader that isolates ABI churn:
+
+```c
+llama_model * llama_model_load_range(
+    const char * path,
+    llama_model_params params,
+    int32_t il_start,
+    int32_t il_end);
+```
+
+Relevant source:
+
+- `include/llama.h:295-331`.
+- `src/llama-model.cpp:2300-2319`.
+
+Adding fields directly to the by-value public parameter structure changes ABI. A project-owned API around internal model state is safer for a pinned worker executable.
+
+### 6.3 Partial weight loading
+
+Current loader behavior already makes a small patch possible:
+
+- Architectures create/register tensors.
+- Backend buffers are allocated only for created tensor contexts.
+- Mmap ranges derive from created tensors.
+- `done_getting_tensors(partial=true)` exists.
+- Data loading only processes tensors present in created contexts.
+
+Relevant source:
+
+- `src/llama-model.cpp:1229-1647`.
+- `src/llama-model-loader.cpp:1054-1572`.
+
+For an initial Llama-family adapter:
+
+- Create token embedding only if `il_start == 0`.
+- Create final norm/output only if `il_end == n_layer`.
+- Create repeating tensors only for `[il_start, il_end)`.
+- Finish model loading with partial mode.
+
+Relevant source:
+
+- `src/models/llama.cpp:34-92`.
+- `src/llama-model.cpp:1490`.
+
+### 6.4 Residual input
+
+Current input machinery supports F32 supplied embeddings:
+
+- `src/llama-graph.h:120-150`.
+- `src/llama-graph.cpp:66-121`.
+- `src/llama-graph.cpp:2151-2229`.
+
+A middle shard needs a dedicated residual input that bypasses token lookup, embedding scaling, LoRA, and padding behavior while still accepting position and sequence IDs for RoPE/KV.
+
+### 6.5 Ranged graph execution
+
+The initial Llama graph loop is in:
+
+- `src/models/llama.cpp:98-247`.
+
+It must:
+
+- Iterate only `[il_start, il_end)`.
+- Preserve all boundary rows for intermediate shards.
+- Avoid final-token-only row pruning except on the actual tail.
+- Run final norm and LM head only on the actual tail.
+
+### 6.6 Boundary output
+
+The network boundary is the unnormalized residual stream after the final owned transformer layer, not llama.cpp's final embedding tensor.
+
+Add a dedicated graph result such as `t_boundary` in:
+
+- `src/llama-graph.h:791-865`.
+- `src/llama-graph.cpp:1190-1251`.
+- `src/llama-context.cpp:2206-2231`.
+
+Existing `llama-ext` layer-input extraction is useful precedent but is explicitly staging/WIP and only populated by some architectures.
+
+### 6.7 Shard-local KV
+
+The standard KV cache already supports a layer filter and compact global-to-local mapping:
+
+- `src/llama-kv-cache.cpp:64-100`.
+- `src/llama-kv-cache.cpp:163-248`.
+- `src/llama-kv-cache.cpp:1210-1327`.
+- `src/llama-model.cpp:2152-2266`.
+
+For a standard Llama adapter, the filter is conceptually:
+
+```cpp
+return il >= il_start && il < il_end;
+```
+
+Do not apply this blindly to hybrid, recurrent, shared-cache, MLA, or other architecture-specific memory implementations.
+
+### 6.8 Route-session mapping
+
+Existing sequence APIs support:
+
+- Sequence removal, copy, keep, and position operations.
+- Sequence state get/set.
+- Partial/on-device state flags.
+
+Relevant source:
+
+- `include/llama.h:720-913`.
+- `src/llama-context.cpp:2900-2968`.
+- `src/llama-context.cpp:3978-4024`.
+
+A shard worker can map `(route_session_id, route_epoch)` to a stable `llama_seq_id` or one isolated context. Because the model contains only filtered layers, sequence state naturally represents shard-local cache.
+
+### 6.9 Scheduler
+
+No scheduler rewrite should be required. The current graph scheduler can allocate a new residual input and boundary output if they are registered correctly.
+
+Relevant source:
+
+- `src/llama-context.cpp:1286-1355`.
+- `src/llama-context.cpp:2324-2459`.
+
+### 6.10 Architecture coupling
+
+A fully generic minimal patch is not plausible:
+
+- The audited tree has roughly 136 model implementation files.
+- More than one hundred contain architecture-specific layer loops.
+- Architectures differ in cache, residual, attention, MoE, recurrent, multimodal, and output behavior.
+
+Generic infrastructure is appropriate for:
+
+- Range validation.
+- Tensor filtering.
+- Residual input/output plumbing.
+- Standard KV filtering.
+- C ABI and worker protocol.
+
+Each supported architecture still needs a reviewed range-aware adapter and real distributed validation. Unsupported architectures must fail closed.
+
+## 7. Selected architecture
+
+```text
+Existing tracker and accounting
+              |
+Existing versioned activation/relay protocol
+              |
+Project-owned standalone C++ GGUF shard worker
+(borrowing or jointly maintaining narrow Nakshatra patch concepts where practical)
+              |
+Pinned llama.cpp fork with small architecture-specific patch series
+              |
+GGUF loader, model graphs, quant kernels, KV, CPU/GPU backends
+```
+
+Nakshatra's working patch and daemon should be reproduced, rebased, and used as source/test evidence before equivalent code is independently designed. Collaboration on narrow upstreamable llama.cpp hooks is preferable to duplicate patch families, but the project should keep its own small pinned fork and standalone worker rather than fork Nakshatra's whole Petals-derived repository. The project-owned boundary remains necessary because Nakshatra's build, control plane, protobuf, sub-GGUF artifact model, single-session daemon, and accounting semantics do not satisfy the full Meshnet contract.
+
+### 7.1 Responsibility split
+
+Python/node agent remains responsible for:
+
+- Tracker registration and heartbeat.
+- Capability proof and admission.
+- Artifact distribution and verification.
+- Route selection and effective start layers.
+- Activation transport and relay.
+- Route sessions, epochs, cache-miss recovery, and process supervision.
+- Work telemetry and accounting.
+
+The C++ worker is responsible for:
+
+- Loading exactly one model/range/recipe.
+- Token embedding on a head shard.
+- Boundary activation input on middle/tail shards.
+- Executing only owned layers.
+- Returning boundary activations on non-tail shards.
+- Final norm/logits/sampling or token output on a tail shard.
+- Local KV/recurrent state by mapped sequence.
+- Exporting health, cache, load, compute, and memory metrics.
+
+### 7.2 Stable worker boundary
+
+Do not expose `ggml_tensor *`, scheduler objects, or llama.cpp structs to Python. Prefer a supervised executable with a small project-owned API, for example:
+
+```text
+load_layer_range
+prefill_tokens
+prefill_activation
+decode_token
+decode_activation
+get_boundary
+get_logits_or_token
+release_session
+export_session
+import_session
+health
+metrics
+```
+
+The wire carries at least:
+
+```text
+protocol version
+request/work ID
+route session ID
+route epoch
+model/artifact fingerprint
+layer begin/end
+effective start layer
+prefill/decode phase
+token start/count
+hidden shape and dtype
+payload length/checksum
+```
+
+### 7.3 Battle-proven transport and concurrency
+
+The implementation program chooses gRPC over HTTP/2 with Protocol Buffers for the standalone Python/C++ Shard data plane rather than inventing a raw socket protocol.
+
+- One long-lived bidirectional stream serves one Route Session Activation Seam.
+- HTTP/2 supplies connection reuse and flow control; gRPC supplies deadlines, cancellation, status, TLS hooks, and generated Python/C++ schemas.
+- Large prefill tensors are split into bounded frames; decode uses a small fast path.
+- Existing relay/WebSocket infrastructure may transport the same protobuf frames as opaque binary when direct gRPC reachability is unavailable.
+- OpenAI-facing HTTP/SSE and existing Tracker APIs remain unchanged.
+
+The public payload is a versioned named-tensor bundle, not one anonymous activation, because architecture boundaries such as vLLM's Llama/Qwen3-MoE pipeline stages can require both `hidden_states` and `residual`.
+
+Concurrency is implemented with local llama.cpp sequences or bounded contexts mapped from `(Route Session ID, route epoch)`. Compatible active decode steps are continuously batched inside each node. This adapts the useful vLLM scheduling concept without importing vLLM's Torch process groups, PagedAttention allocator, or static distributed executor.
+
+Tensor/expert parallel collectives remain confined to a future trusted composite-node or managed-cluster provider. The public volunteer primitive remains contiguous layer Shards.
+
+The benchmark-gated execution plan and Ralph backlog live in [the active distributed-GGUF feature](../../.scratch/distributed-gguf-runtime/README.md).
+
+## 8. Implementation spikes and acceptance gates
+
+### 8.1 Spike 1: reproduce and rebase Nakshatra
+
+Before writing a separate worker, reproduce Nakshatra's dense Llama-family two-worker path and rebase its narrow patch onto the exact current llama.cpp commit selected by this project.
+
+Run two local workers with disjoint sub-GGUF ranges, then compare sub-GGUF loading with a range-aware loader over one shared source artifact.
+
+Acceptance:
+
+- The audited Nakshatra test is independently reproducible rather than accepted from repository claims.
+- Each process maps only its assigned tensors.
+- Each process allocates KV only for assigned layers.
+- Head/middle/tail module ownership is correct.
+- Boundary residuals produce bounded numerical error against whole-model llama.cpp.
+- The patch rebases cleanly onto the pinned current llama.cpp baseline.
+- Reusable worker code is isolated from Nakshatra's Petals-derived and project-specific control plane.
+- Upstream collaboration or shared maintenance is evaluated before duplicating the patch family.
+
+### 8.2 Spike 2: Meshnet protocol and multi-session KV
+
+Place the rebased/Nakshatra-derived worker behind Meshnet's route/session protocol.
+
+Acceptance:
+
+- Multiple route sessions remain isolated through separate `llama_seq_id` values or contexts.
+- Duplicate requests are idempotent and stale route epochs are rejected.
+- Exact source/slice artifact hashes and runtime recipes are checked.
+- Cache miss, release, TTL/LRU eviction, cancellation, and re-prefill are bounded.
+- Killing a worker produces a bounded failure, not a deadlock.
+- Work receipts bind worker identity, request, route, range, artifact, and activation evidence.
+
+### 8.3 Spike 3: heterogeneous two-machine route
+
+Use a tracker-selected two-machine route with real CPU/GPU execution.
+
+Measure:
+
+- Cold and warm model load.
+- Mapped tensor and KV memory per process.
+- Prefill and decode throughput.
+- Activation bytes and boundary latency.
+- Whole-model parity.
+- Process/node failure behavior.
+- Per-node completed work and compute time.
+
+Synthetic tests remain unit coverage, not distributed validation.
+
+### 8.4 Spike 4: Qwen3 MoE adapter
+
+Qwen3 30B-A3B requires an explicit architecture adapter. It must review:
+
+- Expert/router tensor ownership.
+- Top-k routing and MoE graph behavior.
+- Architecture-specific normalization and Q/K normalization.
+- Layer-local expert loading.
+- KV type and layout.
+- Boundary residual placement.
+- Supported HIP/Vulkan/CPU paths.
+
+The adapter should reuse current llama.cpp's Qwen3 graph rather than reconstructing a simplified layer as `llama-gguf` does.
+
+### 8.5 Architecture certification
+
+Each architecture recipe should declare:
+
+- Range-aware graph implementation and version.
+- Head/tail module rules.
+- Cache kind and state support.
+- Boundary dtype/layout.
+- Supported backends.
+- Required runtime commit.
+- Real distributed validation evidence.
+
+Capability admission keeps unsupported combinations registered-but-dark.
+
+## 9. Principal risks
+
+1. **Upstream churn:** llama.cpp model/graph/cache internals change frequently. Keep a small ordered patch series and pinned revisions.
+2. **Architecture coupling:** every architecture adapter needs review and live proof.
+3. **Cache locality:** dynamic rerouting requires replay/checkpoint semantics, not socket retry.
+4. **Boundary compatibility:** shape, dtype, residual point, positions, and cache semantics must match exactly.
+5. **Artifact identity:** exact GGUF/model/tokenizer/runtime fingerprints must be part of capability proof.
+6. **Heterogeneous bottlenecks:** the slowest seam can dominate; placement must use measured end-to-end cost.
+7. **Accounting integrity:** completed-work receipts need route and worker identity and must not trust self-reported latency alone.
+8. **Security:** the worker accepts structured activations, not arbitrary GGML graphs or executable recipes.
+9. **Memory estimates:** weight quantization and KV/state type must both be represented.
+10. **Testing:** every supported lane needs real CPU/GPU tracker-routed acceptance evidence.
+
+## 10. Final conclusion
+
+Distributed GGUF should not be built from zero, but no existing project can be adopted as-is.
+
+The reusable composition is:
+
+- llama.cpp for mature GGUF parsing, architecture graphs, quantized kernels, backend scheduling, and KV/state APIs.
+- prima.cpp for proof and reference implementations of selective local GGUF loading, local layer KV, graph boundaries, and placement ideas.
+- `llama-gguf` for explicit range/protobuf/health/test-organization ideas.
+- Petals and the existing neuron-tai Transformers path for route-session, cache-locality, and recovery semantics.
+- The existing tracker, relay, capability, and accounting systems as the authoritative control plane.
+
+The selected path is a small pinned llama.cpp layer-worker fork with generic shard infrastructure and explicitly certified architecture adapters.

docs/research/vllm-distributed-gguf-assessment.md

@@ -0,0 +1,657 @@
+# vLLM assessment for distributed GGUF inference
+
+Status: Source and documentation assessment
+Last updated: 2026-07-13
+
+## 1. Question
+
+Can vLLM or its GGUF plugin be reused to implement neuron-tai's primary distributed GGUF data plane?
+
+The required data plane is not merely a whole-model serving cluster. It requires independently loaded GGUF layer ranges on heterogeneous nodes, tracker-selected per-request routes, a versioned activation boundary, local route-session cache, dynamic failure handling, relay support, and per-node work attribution.
+
+This assessment also identifies vLLM components and concepts that remain useful even if its distributed runtime is not adopted.
+
+## 2. Sources and snapshots
+
+Source snapshots audited:
+
+```text
+vllm-project/vllm
+commit 107a03ba63e005ff03424fed9c4e6cf551b98bb2
+
+vllm-project/vllm-gguf-plugin
+commit ad209df10bb1856ba53b6663745a831eb7eb09cc
+```
+
+Current official documentation reviewed:
+
+- [GGUF](https://docs.vllm.ai/en/latest/features/quantization/gguf/).
+- [Parallelism and Scaling](https://docs.vllm.ai/en/stable/serving/parallelism_scaling/).
+- [Disaggregated Prefilling](https://docs.vllm.ai/en/latest/features/disagg_prefill/).
+- [Data Parallel Deployment](https://docs.vllm.ai/en/latest/serving/data_parallel_deployment/).
+- [Installation and supported platforms](https://docs.vllm.ai/en/latest/getting_started/installation/).
+- [RFC #39583: Migrate bitsandbytes and GGUF quantization support to an out-of-tree plugin](https://github.com/vllm-project/vllm/issues/39583).
+
+Both source projects are Apache-2.0 licensed.
+
+## 3. Executive decision
+
+### 3.1 Primary distributed GGUF runtime
+
+Do **not** use vLLM as the primary tracker-routed heterogeneous GGUF layer-worker runtime.
+
+Reasons:
+
+- Current GGUF support is an experimental out-of-tree quantization plugin.
+- The plugin converts GGUF tensors into Torch parameters and uses vLLM's PyTorch model implementations; it is not a lightweight llama.cpp/GGML runtime.
+- The plugin builds CUDA/HIP Torch extensions and does not provide CPU, Vulkan, or Metal GGUF execution.
+- vLLM pipeline parallelism assumes one static Torch distributed world with rank/process-group semantics.
+- Pipeline workers communicate with neighboring ranks through Torch distributed or a trusted TCP/device communicator, not the Meshnet HTTP/relay/session protocol.
+- Requests are scheduled as one synchronized engine, not independently routed through tracker-selected workers.
+- Failure of a pipeline rank is a distributed-engine failure, not a route-local cache miss that can be recovered by rebuilding a route.
+- Multi-node guidance requires identical environments and private trusted networking.
+- External vLLM load balancing is across complete data-parallel model replicas, not model-layer workers.
+
+### 3.2 Supported secondary roles
+
+vLLM remains useful in three narrower roles:
+
+1. **Whole-model node backend:** a node proxy can register an independent vLLM server for models/hardware that fit on that deployment. This is an execution recipe, not the core distributed GGUF feature.
+2. **Static managed-cluster backend:** a separately managed, trusted vLLM TP/PP cluster can be exposed as one logical node. The tracker accounts at cluster boundary unless a trusted internal accounting bridge is added.
+3. **Source/design donor:** architecture-aware pipeline boundaries, local layer construction, PagedAttention scheduling, KV connector lifecycles, and telemetry are valuable references.
+
+### 3.3 GGUF kernel reuse
+
+Do not import the vLLM GGUF plugin into the selected llama.cpp shard worker.
+
+The plugin's GGUF kernels are tightly coupled to:
+
+- PyTorch parameters and custom operators.
+- vLLM quantization interfaces.
+- Triton/CUDA/HIP execution.
+- vLLM's model and tensor-parallel parameter layouts.
+- vLLM release internals and monkey-patched plugin registration.
+
+llama.cpp already provides broader GGUF quant coverage, CPU and edge backends, mmap loading, and the desired C/C++ runtime boundary.
+
+## 4. Current GGUF support in vLLM
+
+### 4.1 Support moved out of core
+
+Official documentation states that GGUF support is:
+
+- Highly experimental.
+- Under-optimized.
+- Potentially incompatible with other features.
+- Now provided by the separate `vllm-gguf-plugin`.
+
+The migration RFC gives the maintenance rationale. A vLLM maintainer estimated GGUF usage at roughly 0.1%, described poor performance relative to llama.cpp for batch-size-one consumer-GPU workloads, and identified a large special-case burden: legacy weight-loader branches, model-specific mappings, and roughly 6,000 lines of GGUF CUDA kernels. The out-of-tree plugin preserves availability while allowing vLLM core to prioritize its native production quantization paths. This is evidence that GGUF is a compatibility lane in vLLM, not its primary runtime focus.
+
+At the audited snapshot, the plugin is itself young: version `0.0.4`, with its repository created during the 2026 migration. This is not a reason to reject it as a whole-model backend, but it raises compatibility and maintenance risk for using it as foundational distributed infrastructure.
+
+The plugin registers itself through the `vllm.general_plugins` entry point:
+
+- `vllm-gguf-plugin/pyproject.toml:17-18`.
+- `vllm-gguf-plugin/vllm_gguf_plugin/plugin.py:109-127`.
+
+It patches vLLM engine argument creation and speculative-model probing at runtime:
+
+- `vllm-gguf-plugin/vllm_gguf_plugin/plugin.py:51-97`.
+
+This demonstrates close coupling to vLLM internals and increases upgrade risk.
+
+### 4.2 Dependencies and hardware assumptions
+
+The plugin requires:
+
+- `gguf>=0.17.0`.
+- `vllm`.
+- `torch>=2.9`.
+- CUDA or ROCm toolkit for normal installation.
+
+Evidence:
+
+- `vllm-gguf-plugin/pyproject.toml:1-18`.
+- `vllm-gguf-plugin/README.md:7-13`.
+
+Its compiled extension is created through `torch.utils.cpp_extension.CUDAExtension`. ROCm is handled by compiling the same extension through HIP-compatible tooling:
+
+- `vllm-gguf-plugin/setup.py:28-65`.
+
+This gives a CUDA/ROCm lane, not the CPU/Vulkan/Metal hardware coverage expected from llama.cpp.
+
+### 4.3 Model/config/tokenizer dependency
+
+GGUF is treated as a weight source for a vLLM/Hugging Face model implementation.
+
+The plugin determines model configuration from, in order:
+
+- Explicit `hf_config_path`.
+- A non-GGUF tokenizer path.
+- The remote GGUF repository.
+- The local GGUF parent directory.
+
+Evidence:
+
+- `vllm-gguf-plugin/vllm_gguf_plugin/plugin.py:34-48`.
+- `vllm-gguf-plugin/vllm_gguf_plugin/plugin.py:51-79`.
+
+Official docs recommend using the base Hugging Face tokenizer because converting a tokenizer from GGUF is slow and unstable. If Hugging Face cannot derive the architecture config, users must supply a compatible config path.
+
+This is different from llama.cpp, which treats GGUF metadata, tensors, tokenizer, and architecture implementation as one native runtime artifact.
+
+### 4.4 Artifact resolution
+
+The plugin accepts:
+
+- A local GGUF file.
+- A local directory plus quantization type.
+- A remote `repo_id:quant_type`.
+- A remote `repo_id/filename.gguf`.
+
+Evidence:
+
+- `vllm-gguf-plugin/vllm_gguf_plugin/loader.py:41-67`.
+- `vllm-gguf-plugin/vllm_gguf_plugin/weight_utils.py:18-70`.
+
+It supports split GGUF artifacts named like `-00001-of-00004.gguf`:
+
+- `vllm-gguf-plugin/vllm_gguf_plugin/weights_adapter/default.py:248-264`.
+
+This is file sharding, not model-layer network sharding.
+
+### 4.5 GGUF-to-vLLM name mapping
+
+The default adapter:
+
+1. Maps a GGUF architecture to Hugging Face tensor names through the Python `gguf` package.
+2. Instantiates a Hugging Face model on the `meta` device to obtain its state-dict names.
+3. Creates architecture-specific special mappings for MoE and other variants.
+4. Maps those Hugging Face names into vLLM's packed parameter layout.
+
+Evidence:
+
+- `vllm-gguf-plugin/vllm_gguf_plugin/weights_adapter/default.py:42-231`.
+- Qwen2/Qwen3 MoE mappings: `vllm-gguf-plugin/vllm_gguf_plugin/weights_adapter/default.py:81-98`.
+- vLLM packed Llama mappings: `vllm/model_executor/models/llama.py:344-354`.
+- vLLM packed Qwen3-MoE mappings: `vllm/model_executor/models/qwen3_moe.py:432-446`.
+
+This provides broad model integration by relying on vLLM and Transformers model classes. It does not expose a generic independently executable GGUF layer object.
+
+### 4.6 Loading behavior
+
+`GGUFModelLoader` initializes an ordinary vLLM model on the target Torch device and calls its `load_weights` method with tensors yielded by the adapter:
+
+- `vllm-gguf-plugin/vllm_gguf_plugin/loader.py:78-104`.
+
+The GGUF iterator:
+
+- Opens each GGUF file through `gguf.GGUFReader`.
+- Enumerates every tensor.
+- Converts NumPy-backed data to a Torch tensor.
+- Emits an extra quantization-type parameter for quantized weights.
+
+Evidence:
+
+- `vllm-gguf-plugin/vllm_gguf_plugin/weight_utils.py:73-135`.
+
+Pipeline stages instantiate only their repeating local layers, so vLLM's `AutoWeightsLoader` can skip parameters represented by `PPMissingLayer`. However, the plugin still enumerates the artifact through Python on each worker. This is not the same as selectively registering only a layer range and mmaping only its file regions in llama.cpp.
+
+### 4.7 Quantization types and kernels
+
+The plugin declares support for:
+
+- Unquantized F32, F16, BF16.
+- Q4_0, Q4_1, Q5_0, Q5_1, Q8_0, Q8_1.
+- Q2_K, Q3_K, Q4_K, Q5_K, Q6_K.
+- IQ1_M, IQ1_S, IQ2_XXS, IQ2_XS, IQ2_S, IQ3_XXS, IQ3_S, IQ4_XS, IQ4_NL.
+
+Evidence:
+
+- `vllm-gguf-plugin/vllm_gguf_plugin/quantization/utils.py:46-75`.
+
+Linear execution chooses among:
+
+- Native unquantized matrix multiplication.
+- Quantized matrix-vector kernels for small token counts.
+- Quantized matrix-matrix kernels for larger token counts.
+- Full dequantization followed by Torch matrix multiplication as fallback.
+
+Evidence:
+
+- `vllm-gguf-plugin/vllm_gguf_plugin/quantization/linear.py:34-57`.
+- `vllm-gguf-plugin/vllm_gguf_plugin/quantization/linear.py:79-247`.
+
+There are separate Triton and compiled CUDA/HIP implementations for GEMM, dequantization, embedding, and fused MoE.
+
+These kernels are useful to the vLLM ecosystem but are not portable drop-in components for a llama.cpp/GGML worker.
+
+### 4.8 Test evidence
+
+Plugin tests compare GGUF output against unquantized vLLM output for small models including Qwen2.5, Qwen3 dense, Phi3, GPT-2, StableLM, Gemma3, and OLMoE:
+
+- `vllm-gguf-plugin/tests/test_gguf_generation.py:22-79`.
+- `vllm-gguf-plugin/tests/test_gguf_generation.py:177-226`.
+
+There is a two-GPU tensor-parallel GGUF test for Qwen3-0.6B:
+
+- `vllm-gguf-plugin/tests/test_gguf_generation.py:229-283`.
+
+The core vLLM plugin test also covers Qwen3 dense and OLMoE at TP=1 and TP=2:
+
+- `tests/plugins_tests/gguf/test_gguf_plugin_generate.py:27-132`.
+
+No source test was found for:
+
+- GGUF pipeline parallelism.
+- Multi-node GGUF PP.
+- Mixed hardware.
+- Qwen3 30B-A3B GGUF.
+- Tracker-selected ranges.
+- Independent route sessions or dynamic route recovery.
+
+Kernel tests exercise CUDA tensors and broad quant types. One variable-length batching kernel test is explicitly skipped because the current CUDA kernel does not support that case:
+
+- `vllm-gguf-plugin/tests/test_kernels.py:260-310`.
+
+## 5. vLLM pipeline parallelism
+
+### 5.1 What it gets right
+
+vLLM has a mature architecture-aware pipeline abstraction:
+
+- Model implementations construct only repeating layers assigned to the local PP rank.
+- Missing layers are represented with `PPMissingLayer` placeholders.
+- First-stage token embeddings and last-stage norm/head can be conditionally owned.
+- Non-tail stages return named intermediate tensor bundles.
+- The scheduler and model runner overlap stage communication and execution.
+- Uneven layer-count splits are supported.
+
+This is substantially more mature than the static dense-layer reconstruction in `llama-gguf`.
+
+### 5.2 Static stage partitioning
+
+`make_layers` derives `(start_layer, end_layer)` from PP rank and PP world size, creates real modules only for that interval, and fills all other positions with `PPMissingLayer`:
+
+- `vllm/model_executor/models/utils.py:674-719`.
+
+The default partition is approximately even. A static `VLLM_PP_LAYER_PARTITION` environment variable can override the number of layers per rank:
+
+- `vllm/distributed/utils.py:127-172`.
+
+This can represent uneven contiguous ranges, but it remains a launch-time rank partition:
+
+- One ordered process group.
+- One fixed world size.
+- One static set of adjacent ranks.
+- No per-request tracker route.
+- No arbitrary start/end assignment independent of rank.
+- No overlapping shard ranges or Meshnet effective-start semantics.
+
+### 5.3 Llama stage ownership
+
+For Llama:
+
+- The first rank owns token embedding.
+- A tied output rank may also own embedding.
+- Only the local repeating layers are instantiated.
+- Only the last rank owns final RMSNorm.
+- Only the last rank owns LM head and logits processor.
+
+Evidence:
+
+- `vllm/model_executor/models/llama.py:356-395`.
+- `vllm/model_executor/models/llama.py:400-439`.
+- `vllm/model_executor/models/llama.py:466-503`.
+
+This is a good design reference for explicit head, middle, and tail ownership.
+
+### 5.4 Intermediate tensor bundles
+
+`IntermediateTensors` is a dictionary of named Torch tensors:
+
+- `vllm/sequence.py:10-62`.
+
+For Llama and Qwen3-MoE, the PP boundary includes both:
+
+```text
+hidden_states
+residual
+```
+
+Evidence:
+
+- `vllm/model_executor/models/llama.py:393-395`.
+- `vllm/model_executor/models/llama.py:408-433`.
+- `vllm/model_executor/models/qwen3_moe.py:478-518`.
+
+This is an important architectural lesson. Depending on where the boundary is placed, an exact pipeline stage may require more than one tensor. Meshnet's backend-neutral activation envelope should support a named tensor bundle or deliberately choose a canonical boundary after residual fusion.
+
+A universal single `hidden_states` tensor contract may silently break architecture parity.
+
+### 5.5 Qwen3-MoE stage behavior
+
+The audited Qwen3-MoE implementation supports:
+
+- Q/K normalization.
+- Routed experts.
+- Shared experts.
+- Expert/tensor parallel groups.
+- Stage-local repeating layers.
+- `hidden_states` plus `residual` PP boundaries.
+
+Evidence:
+
+- `vllm/model_executor/models/qwen3_moe.py:130-251`.
+- `vllm/model_executor/models/qwen3_moe.py:254-354`.
+- `vllm/model_executor/models/qwen3_moe.py:357-429`.
+- `vllm/model_executor/models/qwen3_moe.py:432-524`.
+
+However, in this source snapshot, Qwen3-MoE constructs embedding, final norm, and LM head without the first/last-rank guards used by Llama:
+
+- Embedding: `vllm/model_executor/models/qwen3_moe.py:463-471`.
+- Final norm: `vllm/model_executor/models/qwen3_moe.py:477`.
+- LM head/logits: `vllm/model_executor/models/qwen3_moe.py:566-589`.
+
+Repeating layer weights remain stage-local, but endpoint parameters appear replicated. This source observation should be validated in a real PP load before using vLLM memory behavior as a reference for Qwen3 30B-A3B.
+
+### 5.6 Stage communication
+
+The worker receives intermediate tensor dictionaries from the previous PP rank, executes the local model, and asynchronously sends output tensors to the next rank:
+
+- `vllm/v1/worker/gpu_worker.py:1045-1088`.
+
+The PP group:
+
+- Sends metadata through a CPU process group.
+- Sends tensors through Torch distributed device or CPU groups.
+- Defaults destination to the next static rank.
+- Defaults source to the previous static rank.
+- Can optimize TP-group slices with all-gather reconstruction.
+
+Evidence:
+
+- `vllm/distributed/parallel_state.py:960-1053`.
+- `vllm/distributed/parallel_state.py:1055-1149`.
+
+A stateless coordinator variant uses a trusted TCP store and optional device communicator:
+
+- `vllm/distributed/stateless_coordinator.py:300-356`.
+
+This is not a versioned application protocol. It has no Meshnet route session, route epoch, model fingerprint, layer range, request work receipt, relay semantics, authentication, or HTTP cache-miss contract.
+
+### 5.7 Static distributed world
+
+Official vLLM deployment guidance recommends:
+
+- Tensor parallelism within a multi-GPU node.
+- Pipeline parallelism across nodes.
+- Ray or multiprocessing as the distributed executor.
+- Identical model paths, packages, and execution environments across all nodes.
+- High-speed networks such as InfiniBand for cross-node TP.
+- Private networking because distributed traffic is unencrypted and may permit code execution if exposed.
+
+These assumptions are appropriate for one trusted managed cluster, not an open heterogeneous volunteer route.
+
+### 5.8 Cache and scheduling semantics
+
+Each PP stage owns attention modules only for its local layers, so its physical KV is stage-local. However, request scheduling and block assignment are coordinated by one vLLM engine. Cache identity is internal request/block state, not an externally routable `(route_session, route_epoch, model, range)` contract.
+
+A pipeline rank cannot independently accept an arbitrary activation from a tracker and safely infer whether it has the required local cache without building a new application protocol around it.
+
+### 5.9 Failure behavior
+
+vLLM's Ray layer can restart actors and Ray Serve can provide deployment-level fault tolerance, but a live PP engine depends on its process group and synchronized ranks. A lost stage invalidates active distributed execution and local cache for that stage.
+
+No source path was found that dynamically replaces one PP rank during an active generation and resumes from another worker's independently reconstructed local cache. This is fundamentally different from Meshnet's route cache-miss and re-prefill recovery model.
+
+## 6. Other vLLM distributed modes
+
+### 6.1 Tensor parallelism
+
+TP splits tensors and performs collectives within most transformer layers. It assumes tightly coordinated ranks and fast communication. It is a poor fit for independent heterogeneous internet nodes but can be valuable inside one trusted node or managed cluster.
+
+GGUF plugin evidence currently includes a two-GPU TP test. That does not validate PP or volunteer routing.
+
+### 6.2 Expert parallelism
+
+Qwen3-MoE and other MoE models can partition experts across EP ranks. Every forward requires routing and collective coordination across the expert group:
+
+- `vllm/model_executor/models/qwen3_moe.py:130-251`.
+
+This is useful inside a high-speed managed cluster. It should not be confused with assigning whole transformer-layer ranges to independent volunteer nodes.
+
+### 6.3 Data parallelism
+
+vLLM DP replicates complete model weights across independent engine ranks. External load balancing can route HTTP requests to separate complete vLLM deployments, and each replica has an independent KV cache.
+
+This maps cleanly to a whole-model node proxy or managed serving lane, not distributed single-model execution.
+
+### 6.4 Disaggregated prefill
+
+Disaggregated prefill runs separate complete vLLM instances for prefill and decode and transfers KV/results through connector implementations. Official documentation explicitly says it does not improve throughput; it separates TTFT and inter-token-latency tuning.
+
+It is not pipeline layer sharding.
+
+The connector interface is still a useful source of lifecycle ideas:
+
+- Scheduler and worker roles.
+- Request-finished ownership transfer.
+- Asynchronous save/load completion.
+- Per-layer KV save/load hooks.
+- Block IDs and invalid-block reporting.
+- Connector metrics and events.
+- Preemption handling before blocks are overwritten.
+
+Evidence:
+
+- `vllm/distributed/kv_transfer/kv_connector/v1/base.py:1-41`.
+- `vllm/distributed/kv_transfer/kv_connector/v1/base.py:124-229`.
+- `vllm/distributed/kv_transfer/kv_connector/v1/base.py:251-300`.
+- `vllm/v1/worker/kv_connector_model_runner_mixin.py:33-113`.
+
+The API is explicitly experimental and subject to change:
+
+- `vllm/distributed/kv_transfer/kv_connector/v1/base.py:184-200`.
+
+Importing it as a dependency would tightly couple Meshnet cache semantics to vLLM. Reusing its lifecycle concepts in the project-owned cache protocol is safer.
+
+## 7. Hardware portability
+
+vLLM supports multiple platform families through separate builds and platform implementations, including CUDA, ROCm, CPU, XPU, and TPU, with additional ecosystem integrations.
+
+This does **not** imply that one model-parallel vLLM world can mix arbitrary platform types. PP/TP communication and model workers share:
+
+- Torch distributed process groups.
+- A platform-selected device type and communicator.
+- Compatible tensor dtypes and model implementations.
+- Identical package/runtime environments.
+- Synchronization and collective assumptions.
+
+The GGUF plugin itself requires CUDA or ROCm and uses a CUDAExtension/HIP build. It does not provide the CPU/Vulkan/Metal GGUF lane needed for broad volunteer hardware.
+
+Therefore vLLM is multi-platform across separate deployments, not demonstrated as heterogeneous within one distributed model replica.
+
+## 8. Security and trust boundary
+
+Official multi-node documentation warns that distributed traffic is unencrypted and can expose code-execution risk. It recommends a private network.
+
+This alone disqualifies direct exposure of vLLM's Ray/Torch distributed plane to untrusted network participants.
+
+A safe Meshnet node can still supervise vLLM behind a local proxy. The external network sees only the project-owned authenticated and bounded protocol.
+
+## 9. Reuse matrix
+
+| vLLM component | Direct code reuse | Design reuse | Decision |
+|---|---:|---:|---|
+| GGUF Python loader | No | Limited | Whole-model vLLM recipe only |
+| GGUF CUDA/HIP/Triton kernels | No for llama.cpp worker | Performance reference | Keep in vLLM lane |
+| `make_layers` / `PPMissingLayer` | No | Yes | Reference for local stage construction |
+| `IntermediateTensors` | No | Strong yes | Add named tensor-bundle concept to activation ABI |
+| Torch PP communicator | No | Limited | Static trusted cluster only |
+| PagedAttention/block manager | No | Yes | Cache budgeting and metrics concepts |
+| KV connector lifecycle | No | Strong yes | Adapt lifecycle semantics, not dependency |
+| Disaggregated prefill connectors | No | Yes | Reference for async KV transfer/checkpointing |
+| DP external load balancing | Via HTTP proxy | Yes | Whole-model vLLM node lane |
+| Ray executor | No for volunteer mesh | Limited | Managed cluster backend only |
+| Qwen3-MoE model graph | No in C++ worker | Strong yes | Cross-check llama.cpp adapter behavior |
+| Request/KV telemetry | Possibly via proxy | Yes | Map to tracker metrics where trustworthy |
+
+## 10. Implications for native GGUF design
+
+### 10.1 Activation ABI must support tensor bundles
+
+The first GGUF design assumed one BF16 boundary residual. vLLM shows that architecture implementations may naturally expose more than one state tensor, such as:
+
+```text
+hidden_states
+residual
+```
+
+Other architectures may need recurrent state, cross-attention state, or auxiliary routing data.
+
+The backend-neutral envelope should support:
+
+```text
+bundle schema/version
+named tensors
+each tensor's shape, dtype, byte order, compression and checksum
+architecture recipe and boundary point
+```
+
+For each certified architecture, either:
+
+- Define a canonical fused single-residual boundary, or
+- Define an exact named bundle schema.
+
+### 10.2 Separate model artifact from execution recipe
+
+A GGUF artifact fingerprint is insufficient by itself. Capability proof should include:
+
+- Artifact hash.
+- Runtime family and version.
+- Architecture adapter/recipe version.
+- Boundary schema version.
+- Backend and kernel lane.
+- Quantization types actually supported.
+
+This prevents a vLLM GGUF route from being mixed with a llama.cpp GGUF route merely because both opened the same file.
+
+### 10.3 Keep whole-model vLLM as a distinct route kind
+
+A vLLM server can be registered as:
+
+```text
+route_kind = whole_model
+runtime = vllm
+artifact/model identity
+quantization = gguf | awq | gptq | fp8 | ...
+```
+
+It should not claim activation-shard compatibility unless an explicit compatible worker protocol exists.
+
+### 10.4 Managed vLLM cluster as one logical node
+
+A trusted vLLM TP/PP/EP deployment can register as one complete provider. Internal ranks remain invisible to tracker route construction; the cluster owns its own static world, cache, and failure behavior.
+
+### 10.5 Reuse the KV-transfer envelope and lifecycle concepts
+
+vLLM's most portable contribution is not PagedAttention code but its control-plane treatment of external KV movement:
+
+- Protocol versions change when schema, wire format, or memory layout changes.
+- Transfer IDs are distinct from engine request IDs.
+- Metadata carries producer/consumer identity, request, block IDs, topology, block size/layout, and backend details.
+- Compatibility fingerprints include model shape, dtype, KV heads/layers, cache dtype, backend, and allocator mode.
+- Send, receive, completion, abort, abort acknowledgement, and failed-block states are explicit.
+- Failed or unavailable external blocks become a cache miss and local-prefill fallback.
+- Cache ownership can be leased until asynchronous transfer completion before blocks are released.
+
+Evidence:
+
+- `vllm/distributed/kv_transfer/kv_connector/v1/nixl/metadata.py:28-43`.
+- `vllm/distributed/kv_transfer/kv_connector/v1/nixl/metadata.py:46-76`.
+- `vllm/distributed/kv_transfer/kv_connector/v1/nixl/metadata.py:79-139`.
+- `vllm/distributed/kv_transfer/kv_connector/v1/nixl/metadata.py:142-189`.
+- `vllm/distributed/kv_transfer/kv_connector/v1/base.py:453-527`.
+- `vllm/distributed/kv_transfer/kv_connector/v1/base.py:547-566`.
+- `vllm/v1/kv_offload/tiering/p2p/session/protocol.py:167-247`.
+- `vllm/v1/kv_offload/tiering/p2p/manager.py:180-202`.
+
+These small metadata/state-machine patterns can be reimplemented in Meshnet's Python control plane with fields such as:
+
+```text
+schema_version
+request_id
+kv_transfer_id
+producer_worker_id
+consumer_worker_id
+model/cache compatibility fingerprint
+block or token range
+dtype and layout
+route/generation epoch
+transfer state
+```
+
+Do not import vLLM's connector ABCs, block allocator, PagedAttention tensors, or worker classes. They are coupled to vLLM's scheduler, PyTorch/Triton layout, and static runtime.
+
+## 11. Optional validation spikes
+
+These spikes are optional and do not replace the llama.cpp layer-worker plan.
+
+### 11.1 Whole-model vLLM GGUF baseline
+
+On compatible CUDA or ROCm hardware:
+
+- Install a pinned vLLM and plugin pair.
+- Run a small Qwen3 dense GGUF.
+- Compare load time, memory, prefill, decode, and output parity with llama.cpp.
+- Verify whether the target Radeon lane builds the plugin successfully.
+
+This establishes a baseline and potential whole-model recipe.
+
+### 11.2 Static two-stage vLLM PP experiment
+
+Using a non-GGUF or small supported GGUF model:
+
+- Run PP=2 in a trusted local environment.
+- Record each stage's loaded parameter names and memory.
+- Capture actual `IntermediateTensors` names, shapes, and dtypes.
+- Validate local KV allocation by stage.
+- Kill one stage and document engine/request recovery behavior.
+
+This validates design assumptions but does not make vLLM suitable for volunteer routing.
+
+### 11.3 Qwen3 30B-A3B compatibility probe
+
+Before claiming a vLLM GGUF whole-model lane:
+
+- Load the exact target GGUF and base tokenizer/config.
+- Verify fused MoE quantization type support.
+- Test TP=1 and the intended TP/EP configuration.
+- Compare deterministic output to llama.cpp.
+- Measure expert memory, endpoint replication, and ROCm behavior.
+
+No current test in the audited repositories establishes this target combination.
+
+## 12. Final conclusion
+
+vLLM is a strong production serving engine and contains the most mature architecture-aware PyTorch pipeline implementation found in this research. Its stage-local model construction and intermediate tensor bundles are valuable design references.
+
+It is not the correct primary runtime for neuron-tai's distributed GGUF layer mesh because its GGUF path is experimental and GPU-oriented, while its distributed execution assumes one static, trusted, synchronized Torch world.
+
+The architecture decision remains:
+
+```text
+Primary distributed GGUF layer mesh:
+    small pinned llama.cpp layer-worker fork
+    + project-owned activation/session protocol
+    + existing tracker, relay and accounting
+
+Secondary whole-model serving lane:
+    local proxy to vLLM/Ollama/llama-server/etc.
+
+Optional managed-cluster lane:
+    one trusted vLLM TP/PP cluster represented as one logical provider
+```
+
+The strongest new lesson from vLLM is that the distributed activation ABI should be architecture-aware and support named tensor bundles rather than assuming every model can be split at a single anonymous hidden-state tensor.

packages/gateway/meshnet_gateway/server.py

@@ -3,7 +3,6 @@
 import http.server
 import hashlib
 import json
-import os
 from collections import Counter
 from dataclasses import dataclass
 import threading
@@ -62,7 +61,7 @@ class _GatewayHTTPServer(http.server.HTTPServer):
 
 
 class _GatewayHandler(http.server.BaseHTTPRequestHandler):
-    def log_message(self, fmt, *args):  # noqa: suppress request logs in tests
+    def log_message(self, fmt, *args):  # suppress request logs in tests
         pass
 
     def do_GET(self):

packages/node/meshnet_node/activation_compression.py

@@ -99,7 +99,12 @@ def compress_activation(body: bytes, policy: CompressionPolicy) -> CompressionRe
     return CompressionResult(candidate, "zstd", len(body), len(candidate), time.monotonic() - started, "compressed")
 
 
-def decompress_activation(body: bytes, encoding: str | None) -> CompressionResult:
+def decompress_activation(
+    body: bytes,
+    encoding: str | None,
+    *,
+    max_output_bytes: int | None = None,
+) -> CompressionResult:
     """Decode a modern zstd body or preserve a legacy raw body with metrics."""
     started = time.monotonic()
     if not encoding:
@@ -110,8 +115,23 @@ def decompress_activation(body: bytes, encoding: str | None) -> CompressionResul
         import zstandard as zstd
     except ImportError as exc:
         raise ValueError("zstd support is unavailable") from exc
+    if max_output_bytes is not None and max_output_bytes < 0:
+        raise ValueError("max_output_bytes must be non-negative")
     try:
-        raw = zstd.ZstdDecompressor().decompress(body)
+        if max_output_bytes is None:
+            raw = zstd.ZstdDecompressor().decompress(body)
+        else:
+            # Cap both decoder window allocation and bytes read. zstandard's
+            # max_window_size unit is KiB.
+            max_window_kib = max(1024, (max_output_bytes + 1023) // 1024)
+            decompressor = zstd.ZstdDecompressor(max_window_size=max_window_kib)
+            # `decompress(max_output_size=...)` may trust a frame's advertised
+            # content size. A bounded stream read enforces the limit regardless
+            # of frame metadata and detects trailing expansion with one byte.
+            with decompressor.stream_reader(body) as reader:
+                raw = reader.read(max_output_bytes + 1)
+            if len(raw) > max_output_bytes:
+                raise ValueError("zstd activation body exceeds its output limit")
     except zstd.ZstdError as exc:
         raise ValueError("invalid zstd activation body") from exc
     return CompressionResult(raw, "zstd", len(body), len(raw), time.monotonic() - started, "decompressed")

packages/node/meshnet_node/architecture_boundary.py

@@ -0,0 +1,186 @@
+"""Certified architecture adapters for the public TensorBundle boundary.
+
+The adapter is intentionally small: it owns boundary names and endpoint rules,
+not transformer execution. llama.cpp owns local graphs; callers select a
+certified adapter before accepting an activation from another Shard.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from enum import Enum
+import struct
+from typing import Callable, Mapping, Sequence
+
+from .native_protocol import (
+    HIDDEN_STATES,
+    ProtocolError,
+    encode_bundle,
+    encode_tensor,
+    pb,
+    validate_tail_result,
+)
+
+
+class Architecture(str, Enum):
+    DENSE = "dense"
+    MOE = "moe"
+    MLA = "mla"
+
+
+class BoundaryStage(str, Enum):
+    HEAD = "head"
+    MIDDLE = "middle"
+    TAIL = "tail"
+
+
+@dataclass(frozen=True)
+class ProtocolIdentity:
+    request_id: str
+    runtime_recipe_digest: str
+    chat_template_id: str
+    chat_template_version: str
+    reasoning_mode: str
+    architecture: Architecture
+
+
+@dataclass(frozen=True)
+class SamplingParameters:
+    temperature: float
+    top_p: float
+    top_k: int
+    seed: int
+
+
+@dataclass(frozen=True)
+class TailOutput:
+    kind: str
+    value: int | object
+
+    @classmethod
+    def sampled_token(cls, token_id: int) -> "TailOutput":
+        if token_id < 0:
+            raise ProtocolError("sampled token id must be non-negative")
+        return cls("sampled_token", token_id)
+
+
+@dataclass(frozen=True)
+class TypedTailResult:
+    identity: ProtocolIdentity
+    sampling: SamplingParameters
+    output_kind: str
+    message: pb.TailResult
+
+    @property
+    def sampled_token_id(self) -> int | None:
+        return self.message.sampled_token_id if self.output_kind == "sampled_token_id" else None
+
+
+@dataclass(frozen=True)
+class ArchitectureBoundaryAdapter:
+    architecture: Architecture
+    required_names: frozenset[str]
+
+    @property
+    def protocol_architecture(self) -> int:
+        return {
+            Architecture.DENSE: pb.ARCHITECTURE_TYPE_DENSE,
+            Architecture.MOE: pb.ARCHITECTURE_TYPE_MOE,
+            Architecture.MLA: pb.ARCHITECTURE_TYPE_MLA,
+        }[self.architecture]
+
+    def bundle_from_token_ids(
+        self,
+        token_ids: Sequence[int],
+        token_embedding: Callable[[int], Sequence[float]],
+    ):
+        """Head-only embedding entry point; middle/tail never receive IDs."""
+        if self.architecture is not Architecture.DENSE:
+            raise ProtocolError("head token embedding is not certified for this architecture")
+        if not token_ids:
+            raise ProtocolError("head requires at least one token id")
+        rows = [tuple(token_embedding(token)) for token in token_ids]
+        if not rows or not rows[0] or any(len(row) != len(rows[0]) for row in rows):
+            raise ProtocolError("token embedding returned inconsistent hidden widths")
+        payload = struct.pack("<" + "f" * (len(rows) * len(rows[0])), *(x for row in rows for x in row))
+        return self.bundle_from_named_payloads({HIDDEN_STATES: payload}, shape=[1, len(rows), len(rows[0])])
+
+    def bundle_from_named_payloads(
+        self, payloads: Mapping[str, bytes], *, shape: Sequence[int] | None = None
+    ):
+        names = set(payloads)
+        if not self.required_names <= names:
+            missing = sorted(self.required_names - names)
+            raise ProtocolError(f"{self.architecture.value} boundary requires {missing}")
+        tensors = []
+        for name, payload in payloads.items():
+            tensor_shape = list(shape) if name == HIDDEN_STATES and shape else [len(payload) // 4]
+            if len(payload) % 4:
+                raise ProtocolError(f"{name!r} F32 fixture payload is not word aligned")
+            tensors.append(encode_tensor(name, payload, tensor_shape, pb.DTYPE_FLOAT32))
+        return encode_bundle(
+            tensors,
+            architecture=self.protocol_architecture,
+            boundary_point="pre_tail_residual",
+        )
+
+    def input_for(self, stage: BoundaryStage, bundle):
+        """Accept architecture state only after the head embedding boundary."""
+        if stage is BoundaryStage.HEAD:
+            raise ProtocolError("head accepts token ids and owns token embedding")
+        if bundle is None:
+            raise ProtocolError(f"{stage.value} requires a TensorBundle")
+        from .native_protocol import decode_bundle
+
+        payloads = decode_bundle(bundle)
+        if bundle.architecture != self.protocol_architecture:
+            raise ProtocolError("boundary architecture does not match certified adapter")
+        if bundle.boundary_point != "pre_tail_residual":
+            raise ProtocolError("unsupported architecture boundary point")
+        if not self.required_names <= set(payloads):
+            raise ProtocolError(f"{self.architecture.value} boundary requires {sorted(self.required_names)}")
+        return bundle
+
+    def tail_result(
+        self, *, identity: ProtocolIdentity, sampling: SamplingParameters, output: TailOutput
+    ) -> TypedTailResult:
+        if identity.architecture is not self.architecture:
+            raise ProtocolError("tail result architecture does not match certified adapter")
+        if not identity.request_id or not identity.runtime_recipe_digest:
+            raise ProtocolError("tail result requires exact request and recipe identity")
+        if output.kind != "sampled_token":
+            raise ProtocolError("uncertified tail output kind")
+        message = pb.TailResult(
+            identity=pb.RequestRecipeIdentity(
+                request_id=identity.request_id,
+                runtime_recipe_digest=identity.runtime_recipe_digest,
+                chat_template_id=identity.chat_template_id,
+                chat_template_version=identity.chat_template_version,
+                reasoning_mode=identity.reasoning_mode,
+                architecture=self.protocol_architecture,
+            ),
+            sampling=pb.SamplingParameters(
+                temperature=sampling.temperature,
+                top_p=sampling.top_p,
+                top_k=sampling.top_k,
+                seed=sampling.seed,
+                greedy=sampling.temperature == 0.0,
+            ),
+            sampled_token_id=int(output.value),
+        )
+        validate_tail_result(message)
+        return TypedTailResult(identity, sampling, "sampled_token_id", message)
+
+
+_ADAPTERS = {
+    Architecture.DENSE: ArchitectureBoundaryAdapter(Architecture.DENSE, frozenset({HIDDEN_STATES})),
+    Architecture.MOE: ArchitectureBoundaryAdapter(Architecture.MOE, frozenset({HIDDEN_STATES, "router_logits"})),
+    Architecture.MLA: ArchitectureBoundaryAdapter(Architecture.MLA, frozenset({HIDDEN_STATES, "mla_position_state"})),
+}
+
+
+def adapter_for(architecture: Architecture | str) -> ArchitectureBoundaryAdapter:
+    try:
+        return _ADAPTERS[Architecture(architecture)]
+    except (KeyError, ValueError):
+        raise ProtocolError(f"unsupported architecture {architecture!r}") from None

packages/node/meshnet_node/capability.py

@@ -20,6 +20,8 @@ import time
 from dataclasses import dataclass, field
 from typing import Any, Mapping
 
+from .runtime_recipe import CompatibilityFingerprint, ShardIdentity
+
 # Layout of the serialized report. Bump when the JSON shape changes.
 CAPABILITY_SCHEMA_VERSION = 1
 
@@ -330,7 +332,16 @@ def _as_mapping(data: Any, field_name: str) -> Mapping[str, Any]:
 
 @dataclass(frozen=True)
 class CapabilityReport:
-    """One node's validated (or failed) model/shard/recipe/backend combination."""
+    """One node's validated (or failed) model/shard/recipe/backend combination.
+
+    `identity` is the exact DGR-003 artifact/runtime-recipe block: the separated
+    numerical axes and the compatibility fingerprint derived from them. It is
+    optional and additive — a node that predates DGR-003 presents none, and the
+    tracker falls back to the coarse label comparison it has always done
+    (ADR-0023's compat rollout). A node that *does* present one is held to it:
+    the tracker re-derives the fingerprint and refuses a report whose claim does
+    not match its own derivation.
+    """
 
     model: ModelIdentity
     shard: ShardRange
@@ -341,6 +352,7 @@ class CapabilityReport:
     duration_ms: int
     diagnostics: tuple[str, ...] = ()
     schema_version: int = CAPABILITY_SCHEMA_VERSION
+    identity: ShardIdentity | None = None
 
     def __post_init__(self) -> None:
         if self.status not in VALID_STATUSES:
@@ -360,6 +372,11 @@ class CapabilityReport:
     def passed(self) -> bool:
         return self.status == STATUS_PASSED
 
+    @property
+    def fingerprint(self) -> CompatibilityFingerprint | None:
+        """The exact compatibility fingerprint, when this node declares one."""
+        return None if self.identity is None else self.identity.fingerprint
+
     def identity_key(self) -> tuple[str, int, int, str, str, str, str]:
         """The tuple a consumer must match to reuse this proof.
 
@@ -380,7 +397,7 @@ class CapabilityReport:
         return max(0.0, (time.time() if now is None else now) - self.validated_at)
 
     def to_dict(self) -> dict:
-        return {
+        doc = {
             "schema_version": self.schema_version,
             "model": self.model.to_dict(),
             "shard": self.shard.to_dict(),
@@ -391,6 +408,9 @@ class CapabilityReport:
             "duration_ms": self.duration_ms,
             "diagnostics": list(self.diagnostics),
         }
+        if self.identity is not None:
+            doc["identity"] = self.identity.to_dict()
+        return doc
 
     def to_json(self, indent: int | None = None) -> str:
         return json.dumps(self.to_dict(), indent=indent, sort_keys=True)
@@ -417,6 +437,7 @@ class CapabilityReport:
         ):
             raise CapabilityReportError("'validated_at' must be a Unix timestamp")
 
+        raw_identity = doc.get("identity")
         return cls(
             schema_version=schema_version,
             model=ModelIdentity.from_dict(doc.get("model")),
@@ -427,6 +448,9 @@ class CapabilityReport:
             validated_at=float(validated_at),
             duration_ms=_require_int(doc.get("duration_ms"), "duration_ms", 0),
             diagnostics=sanitize_diagnostics(doc.get("diagnostics")),
+            identity=(
+                None if raw_identity is None else ShardIdentity.from_dict(raw_identity)
+            ),
         )
 
     @classmethod
@@ -461,12 +485,14 @@ def build_capability_report(
     diagnostics: Any = None,
     validated_at: float | None = None,
     environ: Mapping[str, str] | None = None,
+    identity: ShardIdentity | None = None,
 ) -> CapabilityReport:
     """Assemble a report from flat validation results.
 
     `model_config` may be the loaded config mapping (hashed into a fingerprint)
     or an already-computed ``sha256:…`` string. `validated_at` defaults to now,
-    so callers that need determinism pass it explicitly.
+    so callers that need determinism pass it explicitly. `identity` is the exact
+    DGR-003 artifact/recipe block, when the backend can state one.
     """
     return CapabilityReport(
         model=ModelIdentity(
@@ -491,4 +517,5 @@ def build_capability_report(
         validated_at=time.time() if validated_at is None else validated_at,
         duration_ms=duration_ms,
         diagnostics=sanitize_diagnostics(diagnostics, environ),
+        identity=identity,
     )

packages/node/meshnet_node/cli.py

@@ -144,7 +144,7 @@ def _cmd_default(args) -> int:
             print("\nSetup cancelled.")
             return 1
         save_config(cfg)
-        print(f"\nConfig saved to ~/.config/meshnet/config.json\n")
+        print("\nConfig saved to ~/.config/meshnet/config.json\n")
 
     # Apply CLI overrides on top of saved config
     overrides: dict = {}
@@ -198,7 +198,7 @@ def _cmd_default(args) -> int:
 
 def _cmd_models(args) -> int:
     """List curated models (with optional HF Hub browse)."""
-    from .wizard import print_models_table, _browse_hf_interactive
+    from .wizard import print_models_table
 
     if args.browse:
         from .model_catalog import browse_hf_hub

packages/node/meshnet_node/dashboard.py

@@ -5,7 +5,6 @@ from __future__ import annotations
 import os
 import sys
 import time
-from collections import deque
 from typing import TYPE_CHECKING
 
 if TYPE_CHECKING:
@@ -114,7 +113,7 @@ def run_dashboard(node, config: dict, start_time: float) -> None:
         return
 
     try:
-        from rich.live import Live  # type: ignore[import]
+        from rich.live import Live  # type: ignore[import]  # noqa: F401
 
         _run_rich_dashboard(node, config, start_time)
     except ImportError:
@@ -126,7 +125,6 @@ def _build_rich_renderable(
 ):
     from rich.table import Table  # type: ignore[import]
     from rich.panel import Panel  # type: ignore[import]
-    from rich.columns import Columns  # type: ignore[import]
     from rich.text import Text  # type: ignore[import]
 
     uptime = time.monotonic() - start_time
@@ -178,8 +176,8 @@ def _build_rich_renderable(
         f"Tokens/sec   {tps_bar}  {tps:.1f} t/s (EMA)",
         f"Requests     {req_count:,} served",
         f"Success      {stats['success_rate']:.1f}%  failed {stats['failed_requests']:,}  queue {stats['queue_depth']}",
-        f"Peers        0 connected  (gossip: US-017)",
-        f"TAI earned   0.00 TAI  (payments: US-006)",
+        "Peers        0 connected  (gossip: US-017)",
+        "TAI earned   0.00 TAI  (payments: US-006)",
         f"Uptime       {_format_uptime(uptime)}",
         "",
         "[q] quit   [c] compact view",

packages/node/meshnet_node/doctor.py

@@ -36,6 +36,7 @@ from .capability import (
     CapabilityReport,
     build_capability_report,
 )
+from .native_backend import NativeWorkerBackendAdapter
 from .recipe_manifest import (
     DEFAULT_RECIPE_ID,
     Recipe,
@@ -449,11 +450,9 @@ def _validate_recipe(
     category: str | None = None
     error: BaseException | None = None
     diagnostics: list[str] = []
-    detail: dict = {}
-
     try:
         backend = load_backend(selection, recipe)
-        detail = probe_forward(backend)
+        probe_forward(backend)
     except DoctorError as exc:
         category, error = exc.category, exc
         diagnostics = [str(exc), exc.hint]
@@ -464,23 +463,48 @@ def _validate_recipe(
     duration_ms = int((time.monotonic() - started) * 1000)
 
     device = _backend_device(backend, selection)
+    # Only the native adapter has an authoritative immutable GGUF report and
+    # deployment pin.  The Transformers path deliberately remains dark: a
+    # model/config fingerprint is not an exact ArtifactIdentity.
+    identity = backend.identity if isinstance(backend, NativeWorkerBackendAdapter) else None
+    model_id = selection.model_id if identity is None else identity.artifact.artifact_id
+    shard_start = selection.shard_start if identity is None else identity.shard_start
+    shard_end = selection.shard_end if identity is None else identity.shard_end - 1
+    recipe_id = recipe.id if identity is None else identity.recipe.recipe_id
+    recipe_version = recipe.version if identity is None else identity.recipe.recipe_version
+    catalogue_version = (
+        manifest.catalogue_version if identity is None else identity.recipe.catalogue_version
+    )
+    backend_id = recipe.backend_id if identity is None else identity.recipe.backend_id
+    quantization = (
+        selection.quantization if identity is None else identity.recipe.weight_quantization
+    )
+    runtime = _runtime_versions()
+    model_config = _model_config(backend)
+    revision = None
+    if identity is not None:
+        revision = identity.artifact.revision
+        model_config = "sha256:" + identity.artifact.architecture_digest
+        runtime = {**runtime, "native_runtime": identity.recipe.runtime_version}
     report = build_capability_report(
-        model_id=selection.model_id,
-        shard_start=selection.shard_start,
-        shard_end=selection.shard_end,
-        recipe_id=recipe.id,
-        recipe_version=recipe.version,
-        catalogue_version=manifest.catalogue_version,
-        backend_id=recipe.backend_id,
+        model_id=model_id,
+        shard_start=shard_start,
+        shard_end=shard_end,
+        recipe_id=recipe_id,
+        recipe_version=recipe_version,
+        catalogue_version=catalogue_version,
+        backend_id=backend_id,
         device=device,
         device_name=_backend_device_name(device),
-        quantization=selection.quantization,
-        runtime=_runtime_versions(),
-        model_config=_model_config(backend),
+        quantization=quantization,
+        runtime=runtime,
+        revision=revision,
+        model_config=model_config,
         status=STATUS_FAILED if category else STATUS_PASSED,
         duration_ms=duration_ms,
         diagnostics=[d for d in diagnostics if d] or None,
         validated_at=clock(),
+        identity=identity,
     )
     if category:
         return RecipeResult(

packages/node/meshnet_node/glm_alpha/__init__.py

@@ -0,0 +1,123 @@
+"""The locked GLM-5.2 Max alpha target: identity, resource plan, and acceptance contract.
+
+Three files, three jobs:
+
+- :mod:`~meshnet_node.glm_alpha.manifest` — *is this the exact artifact?* Pinned
+  repository revisions, six shard digests, and the architecture-critical config
+  snapshot they must agree with.
+- :mod:`~meshnet_node.glm_alpha.planner` — *can this route hold it?* Deterministic
+  memory, KV, and seam arithmetic over the exact artifact bytes, counting unified
+  memory once.
+- :mod:`~meshnet_node.glm_alpha.contract` — *what would have counted as success?*
+  The acceptance thresholds, locked before the target runs and digest-bound so a
+  later change cannot be silent.
+
+Nothing here downloads, loads, or executes a model. This package is the contract
+DGR-018, DGR-019, and DGR-020 are judged against.
+"""
+
+from __future__ import annotations
+
+from .contract import (
+    ALPHA_CONTRACT_ID,
+    ALPHA_CONTRACT_SCHEMA_VERSION,
+    VERDICT_ALPHA,
+    VERDICT_STOP,
+    AlphaContract,
+    AlphaContractError,
+    compute_contract_digest,
+    load_alpha_contract,
+    parse_alpha_contract,
+    require_contract_target,
+    seal_contract,
+)
+from .manifest import (
+    ALPHA_QUANTIZATION,
+    ALPHA_SHARD_COUNT,
+    ArchitectureSnapshot,
+    GlmTargetError,
+    Shard,
+    TargetManifest,
+    canonical_sha256,
+    load_architecture_snapshot,
+    load_target_manifest,
+    parse_architecture_snapshot,
+    parse_target_manifest,
+    require_pinned_target,
+)
+from .planner import (
+    AGGREGATE_HARD_FIT_FLOOR_GIB,
+    ALPHA_CONTEXT_TOKENS,
+    ALPHA_KV_DTYPE,
+    MIN_LINK_RATE_GBPS,
+    PLACEMENT_IMBALANCE_FACTOR,
+    RECOMMENDED_LINK_RATE_GBPS,
+    RESERVE_FLOOR_GIB,
+    RESERVE_FRACTION,
+    NodeMemory,
+    ResourcePlanError,
+    RouteFit,
+    SeamPlan,
+    TopologyPlan,
+    kv_bytes,
+    plan_all_tiers,
+    plan_route,
+    plan_seams,
+    plan_topology,
+)
+
+
+def load_locked_target() -> tuple[AlphaContract, TargetManifest, ArchitectureSnapshot]:
+    """Load and cross-bind the packaged alpha contract, manifest, and snapshot."""
+
+    contract = load_alpha_contract()
+    manifest = load_target_manifest()
+    snapshot = load_architecture_snapshot()
+    require_contract_target(contract, manifest, snapshot)
+    return contract, manifest, snapshot
+
+
+__all__ = [
+    "AGGREGATE_HARD_FIT_FLOOR_GIB",
+    "ALPHA_CONTEXT_TOKENS",
+    "ALPHA_CONTRACT_ID",
+    "ALPHA_CONTRACT_SCHEMA_VERSION",
+    "ALPHA_KV_DTYPE",
+    "ALPHA_QUANTIZATION",
+    "ALPHA_SHARD_COUNT",
+    "MIN_LINK_RATE_GBPS",
+    "PLACEMENT_IMBALANCE_FACTOR",
+    "RECOMMENDED_LINK_RATE_GBPS",
+    "RESERVE_FLOOR_GIB",
+    "RESERVE_FRACTION",
+    "VERDICT_ALPHA",
+    "VERDICT_STOP",
+    "AlphaContract",
+    "AlphaContractError",
+    "ArchitectureSnapshot",
+    "GlmTargetError",
+    "NodeMemory",
+    "ResourcePlanError",
+    "RouteFit",
+    "SeamPlan",
+    "Shard",
+    "TargetManifest",
+    "TopologyPlan",
+    "canonical_sha256",
+    "compute_contract_digest",
+    "kv_bytes",
+    "load_alpha_contract",
+    "load_architecture_snapshot",
+    "load_locked_target",
+    "load_target_manifest",
+    "parse_alpha_contract",
+    "parse_architecture_snapshot",
+    "parse_target_manifest",
+    "plan_all_tiers",
+    "plan_route",
+    "plan_seams",
+    "plan_topology",
+    "require_contract_target",
+    "require_pinned_target",
+    "seal_contract",
+]

packages/node/meshnet_node/glm_alpha/contract.py

@@ -0,0 +1,359 @@
+"""The immutable GLM-5.2 Max alpha acceptance contract.
+
+The contract exists to answer one question that cannot be answered honestly after
+the fact: *what would have counted as success?*
+
+Its thresholds are locked before the target ever runs (DGR-017), and DGR-020 reads
+them back to publish an ``alpha`` or ``stop`` verdict. The whole point is that the
+gap between those two moments is where a threshold quietly becomes "0.1 tokens/sec
+was always the goal". So the document carries ``contract_sha256`` over its canonical content, while the
+approved v1 digest is pinned independently in code. :func:`load_alpha_contract`
+recomputes the self-digest and then requires that trusted pre-execution digest.
+Changing a threshold and re-sealing under the same identity is rejected; an
+amendment requires a new supported contract identity under human review.
+
+The parsed contract recursively freezes nested mappings and sequences. Thresholds
+therefore cannot change between verification and use, and :meth:`AlphaContract.to_dict`
+returns an isolated mutable copy for diagnostics and tests.
+"""
+
+from __future__ import annotations
+
+import json
+import re
+from dataclasses import dataclass
+from importlib.resources import files
+from pathlib import Path
+from types import MappingProxyType
+from typing import Any, Mapping
+
+from .manifest import (
+    ALPHA_QUANTIZATION,
+    ALPHA_SHARD_COUNT,
+    ArchitectureSnapshot,
+    GlmTargetError,
+    TargetManifest,
+    canonical_sha256,
+)
+
+ALPHA_CONTRACT_SCHEMA_VERSION = 1
+ALPHA_CONTRACT_VERSION = 1
+ALPHA_CONTRACT_ID = "glm-5.2-max-alpha/v1"
+ALPHA_CONTRACT_V1_SHA256 = "aab23220280c053a3c14ff559df3cb5c9e1bf7f0f7188c6519e2e9d9ad036ed9"
+
+_CONTRACT_RESOURCE = "alpha-contract.json"
+
+DIGEST_FIELD = "contract_sha256"
+
+VERDICT_ALPHA = "alpha"
+VERDICT_STOP = "stop"
+
+# Every section the roadmap's acceptance matrix (section 5) locks. A contract that
+# omits one is not a weaker contract, it is an unreviewable one.
+REQUIRED_SECTIONS: tuple[str, ...] = (
+    "identity_and_fit",
+    "semantic_correctness",
+    "target_run",
+    "performance",
+    "reliability",
+    "storage",
+)
+
+
+class AlphaContractError(GlmTargetError):
+    """Raised when the alpha contract is missing, malformed, or has been mutated."""
+
+
+def contract_signing_payload(document: Mapping[str, Any]) -> dict:
+    """The contract content the digest covers: everything except the digest itself."""
+    unsigned = dict(document)
+    unsigned.pop(DIGEST_FIELD, None)
+    return unsigned
+
+
+def compute_contract_digest(document: Mapping[str, Any]) -> str:
+    """SHA-256 over the canonical contract content."""
+    return canonical_sha256(contract_signing_payload(_thaw_json(document)))
+
+
+def _freeze_json(value: Any) -> Any:
+    if isinstance(value, Mapping):
+        return MappingProxyType({str(key): _freeze_json(item) for key, item in value.items()})
+    if isinstance(value, list):
+        return tuple(_freeze_json(item) for item in value)
+    return value
+
+
+def _thaw_json(value: Any) -> Any:
+    if isinstance(value, Mapping):
+        return {str(key): _thaw_json(item) for key, item in value.items()}
+    if isinstance(value, tuple):
+        return [_thaw_json(item) for item in value]
+    return value
+
+
+@dataclass(frozen=True)
+class AlphaContract:
+    """A locked, digest-bound alpha acceptance contract."""
+
+    schema_version: int
+    contract_version: int
+    contract_id: str
+    locked_at: str
+    locked_by: str
+    target: Mapping[str, Any]
+    sections: Mapping[str, Mapping[str, Any]]
+    verdicts: tuple[str, ...]
+    amendment_policy: str
+    digest: str
+    raw: Mapping[str, Any]
+    source: str = "<memory>"
+
+    def section(self, name: str) -> Mapping[str, Any]:
+        if name not in self.sections:
+            raise AlphaContractError(f"contract section {name!r} is missing from {self.source}")
+        return self.sections[name]
+
+    def threshold(self, section: str, key: str) -> Any:
+        block = self.section(section)
+        if key not in block:
+            raise AlphaContractError(
+                f"threshold {section}.{key} is not locked in {self.source}; an unlocked "
+                "threshold cannot be used to judge a result"
+            )
+        return block[key]
+
+    def to_dict(self) -> dict:
+        return _thaw_json(self.raw)
+
+
+def parse_alpha_contract(data: Any, source: str = "<memory>") -> AlphaContract:
+    """Validate a contract document and verify it has not been mutated since locking."""
+    if not isinstance(data, Mapping):
+        raise AlphaContractError(f"contract root in {source} must be a JSON object")
+
+    schema_version = data.get("schema_version")
+    if (
+        not isinstance(schema_version, int)
+        or isinstance(schema_version, bool)
+        or schema_version != ALPHA_CONTRACT_SCHEMA_VERSION
+    ):
+        raise AlphaContractError(
+            f"{source} declares alpha-contract schema version {schema_version!r}, but this "
+            f"node reads version {ALPHA_CONTRACT_SCHEMA_VERSION}"
+        )
+
+    contract_version = data.get("contract_version")
+    if (
+        not isinstance(contract_version, int)
+        or isinstance(contract_version, bool)
+        or contract_version != ALPHA_CONTRACT_VERSION
+    ):
+        raise AlphaContractError(
+            f"{source} declares contract version {contract_version!r}, but this node "
+            f"reads version {ALPHA_CONTRACT_VERSION}"
+        )
+
+    contract_id = data.get("contract_id")
+    if contract_id != ALPHA_CONTRACT_ID:
+        raise AlphaContractError(
+            f"{source} declares contract_id {contract_id!r}, but this node is locked "
+            f"to {ALPHA_CONTRACT_ID!r}"
+        )
+
+    for field in ("locked_at", "locked_by"):
+        value = data.get(field)
+        if not isinstance(value, str) or not value.strip():
+            raise AlphaContractError(f"{source} must carry a non-empty {field}")
+
+    declared = data.get(DIGEST_FIELD)
+    if not isinstance(declared, str) or not declared:
+        raise AlphaContractError(
+            f"{source} carries no {DIGEST_FIELD}; an unsealed contract cannot prove it "
+            "predates the results it judges"
+        )
+
+    computed = compute_contract_digest(data)
+    if computed != declared:
+        raise AlphaContractError(
+            f"{source} has been modified since it was locked: its content hashes to "
+            f"{computed}, but it declares {declared}. Alpha thresholds are locked before "
+            "target execution and may not be weakened afterwards. To change them, open a "
+            "new contract_id under human review; do not edit this one."
+        )
+
+    if not data.get("locked_before_target_execution"):
+        raise AlphaContractError(
+            f"{source} does not assert locked_before_target_execution; a contract written "
+            "after the results are known is not a contract"
+        )
+
+    missing = [name for name in REQUIRED_SECTIONS if not isinstance(data.get(name), Mapping)]
+    if missing:
+        raise AlphaContractError(
+            f"{source} is missing locked acceptance section(s) {missing}"
+        )
+
+    verdicts = data.get("verdicts")
+    if not isinstance(verdicts, list) or sorted(verdicts) != sorted([VERDICT_ALPHA, VERDICT_STOP]):
+        raise AlphaContractError(
+            f"{source} must offer exactly the verdicts {[VERDICT_ALPHA, VERDICT_STOP]}; a "
+            "third outcome is how 'it loaded' becomes a pass"
+        )
+
+    target = data.get("target")
+    if not isinstance(target, Mapping):
+        raise AlphaContractError(f"{source} is missing its 'target' block")
+    for field in (
+        "source_repo_id",
+        "source_revision",
+        "gguf_repo_id",
+        "gguf_revision",
+        "quantization",
+        "target_manifest_sha256",
+        "architecture_snapshot_sha256",
+        "reasoning_effort",
+    ):
+        if not target.get(field):
+            raise AlphaContractError(f"{source} target block is missing {field!r}")
+    for field in ("source_revision", "gguf_revision"):
+        if not re.fullmatch(r"[0-9a-f]{40}", str(target[field])):
+            raise AlphaContractError(f"{source} target.{field} is not a full commit revision")
+    for field in ("target_manifest_sha256", "architecture_snapshot_sha256"):
+        if not re.fullmatch(r"[0-9a-f]{64}", str(target[field])):
+            raise AlphaContractError(f"{source} target.{field} is not a SHA-256 digest")
+    if target["quantization"] != ALPHA_QUANTIZATION:
+        raise AlphaContractError(
+            f"{source} targets {target['quantization']!r}, not locked alpha quantization "
+            f"{ALPHA_QUANTIZATION!r}"
+        )
+    if target["reasoning_effort"] != "max":
+        raise AlphaContractError(f"{source} must lock reasoning_effort='max'")
+    shard_count = target.get("shard_count")
+    if (
+        not isinstance(shard_count, int)
+        or isinstance(shard_count, bool)
+        or shard_count != ALPHA_SHARD_COUNT
+    ):
+        raise AlphaContractError(
+            f"{source} target.shard_count must be exactly {ALPHA_SHARD_COUNT}"
+        )
+    total_bytes = target.get("total_bytes")
+    if not isinstance(total_bytes, int) or isinstance(total_bytes, bool) or total_bytes <= 0:
+        raise AlphaContractError(f"{source} target.total_bytes must be a positive integer")
+
+    amendment_policy = data.get("amendment_policy")
+    if not isinstance(amendment_policy, str) or not amendment_policy.strip():
+        raise AlphaContractError(f"{source} must state its amendment policy")
+
+    if declared != ALPHA_CONTRACT_V1_SHA256:
+        raise AlphaContractError(
+            f"{source} is a re-sealed mutation of {ALPHA_CONTRACT_ID}: digest "
+            f"{declared} does not match the trusted pre-execution digest "
+            f"{ALPHA_CONTRACT_V1_SHA256}. An amendment requires a new supported "
+            "contract identity under human review."
+        )
+
+    frozen = _freeze_json(data)
+
+    return AlphaContract(
+        schema_version=schema_version,
+        contract_version=contract_version,
+        contract_id=contract_id,
+        locked_at=str(data["locked_at"]),
+        locked_by=str(data["locked_by"]),
+        target=frozen["target"],
+        sections=MappingProxyType({name: frozen[name] for name in REQUIRED_SECTIONS}),
+        verdicts=tuple(verdicts),
+        amendment_policy=amendment_policy,
+        digest=declared,
+        raw=frozen,
+        source=source,
+    )
+
+
+def require_contract_target(
+    contract: AlphaContract,
+    manifest: TargetManifest,
+    snapshot: ArchitectureSnapshot,
+) -> None:
+    """Bind the sealed contract to the exact manifest and architecture snapshot.
+
+    Repository revisions alone do not bind shard LFS objects or derived architecture
+    semantics. Call this before planning, admission, download, or execution.
+    """
+
+    expected = contract.target
+    actual = {
+        "source_repo_id": manifest.source_repo_id,
+        "source_revision": manifest.source_revision,
+        "gguf_repo_id": manifest.gguf_repo_id,
+        "gguf_revision": manifest.gguf_revision,
+        "quantization": manifest.quantization,
+        "shard_count": len(manifest.shards),
+        "total_bytes": manifest.total_bytes,
+        "target_manifest_sha256": manifest.digest,
+        "architecture_snapshot_sha256": snapshot.digest,
+    }
+    if snapshot.source_repo_id != manifest.source_repo_id:
+        raise AlphaContractError(
+            "architecture snapshot repository does not match the target manifest repository"
+        )
+    if snapshot.source_revision != manifest.source_revision:
+        raise AlphaContractError(
+            "architecture snapshot revision does not match the target manifest revision"
+        )
+    mismatches = {
+        key: (expected.get(key), value)
+        for key, value in actual.items()
+        if expected.get(key) != value
+    }
+    if mismatches:
+        details = ", ".join(
+            f"{key}: locked={locked!r}, actual={value!r}"
+            for key, (locked, value) in sorted(mismatches.items())
+        )
+        raise AlphaContractError(f"target documents do not match the sealed contract: {details}")
+
+
+def load_alpha_contract(path: Path | None = None) -> AlphaContract:
+    """Load the packaged alpha contract, or one at ``path``."""
+    if path is not None:
+        source = str(path)
+        try:
+            raw = path.read_text(encoding="utf-8")
+        except OSError as exc:
+            raise AlphaContractError(f"cannot read {source}: {exc.strerror or exc}") from exc
+    else:
+        source = f"packaged {_CONTRACT_RESOURCE}"
+        try:
+            raw = (
+                files("meshnet_node.glm_alpha")
+                .joinpath("data", _CONTRACT_RESOURCE)
+                .read_text(encoding="utf-8")
+            )
+        except (OSError, FileNotFoundError, ModuleNotFoundError) as exc:
+            raise AlphaContractError(
+                f"{source} is missing from this node installation ({type(exc).__name__})"
+            ) from exc
+
+    try:
+        data = json.loads(raw)
+    except json.JSONDecodeError as exc:
+        raise AlphaContractError(
+            f"{source} is not valid JSON: {exc.msg} at line {exc.lineno} column {exc.colno}"
+        ) from exc
+
+    return parse_alpha_contract(data, source=source)
+
+
+def seal_contract(document: Mapping[str, Any]) -> dict:
+    """Return the document with a freshly computed digest.
+
+    This is the only supported way to produce a contract file. It is deliberately
+    *not* called at load time: sealing on load would turn every mutation into a
+    valid contract, which is precisely the property the digest exists to deny.
+    """
+    sealed = dict(document)
+    sealed[DIGEST_FIELD] = compute_contract_digest(document)
+    return sealed

packages/node/meshnet_node/glm_alpha/data/alpha-contract.json

@@ -0,0 +1,122 @@
+{
+  "schema_version": 1,
+  "contract_version": 1,
+  "contract_id": "glm-5.2-max-alpha/v1",
+  "locked_at": "2026-07-13",
+  "locked_by": "DGR-017",
+  "locked_before_target_execution": true,
+  "target": {
+    "source_repo_id": "zai-org/GLM-5.2",
+    "source_revision": "b4734de4facf877f85769a911abafc5283eab3d9",
+    "gguf_repo_id": "unsloth/GLM-5.2-GGUF",
+    "gguf_revision": "abc55e72527792c6e77069c99b4cb7de16fa9f23",
+    "quantization": "UD-IQ1_S",
+    "shard_count": 6,
+    "total_bytes": 216715360960,
+    "target_manifest_sha256": "0b6aed04479d204902bb64c0203f1a46cab26a47b378ecccf85237b63f6c1962",
+    "architecture_snapshot_sha256": "253fbd94b06b42acc4724ec2c7f33914e2d4cc43f54a36dff6af19a80ae6ceb1",
+    "reasoning_effort": "max"
+  },
+  "identity_and_fit": {
+    "require_exact_revisions": true,
+    "require_all_shard_sha256": true,
+    "require_per_node_owned_tensor_report": true,
+    "require_owned_tensor_union_equals_inventory": true,
+    "max_unintended_tensor_overlap": 0,
+    "require_no_single_node_can_admit_complete_recipe": true,
+    "min_node_reserve_fraction": 0.2,
+    "min_node_reserve_gib": 8.0,
+    "require_measured_peak_scratch_inside_reserve": true,
+    "forbid_swap": true,
+    "forbid_overcommit": true,
+    "forbid_mmap_only_fit_claim": true,
+    "forbid_double_counted_unified_memory": true,
+    "aggregate_hard_fit_floor_gib": 224.0,
+    "aggregate_floor_class": "experimental_hard_fit_floor",
+    "recommended_topologies": [
+      "5x64GiB",
+      "3x96GiB",
+      "3x128GiB"
+    ],
+    "arithmetic_minimum_requires_measured_placement_evidence": true
+  },
+  "semantic_correctness": {
+    "require_active_moe_routing": true,
+    "require_active_shared_expert": true,
+    "require_active_dsa_lightning_indexer": true,
+    "require_active_sparse_attention": true,
+    "require_active_indexshare_full_and_shared": true,
+    "dense_attention_fallback_satisfies_alpha": false,
+    "require_rendered_reasoning_effort_marker": "<|system|>Reasoning Effort: Max",
+    "f32_seam_fixture_exact_match_tokens": 32,
+    "f32_seam_fixture_requires_exact_match": true,
+    "min_greedy_token_agreement": 0.9,
+    "min_mean_state_cosine_similarity": 0.999,
+    "forbid_nonfinite_tensors": true,
+    "require_fail_closed_on_fingerprint_mismatch": true
+  },
+  "target_run": {
+    "context_tokens": 16384,
+    "kv_dtype": "Q8_0",
+    "concurrency": 1,
+    "prompt_lane_tokens": 4096,
+    "min_output_tokens": 512,
+    "min_output_tokens_with_natural_eos": 128,
+    "require_same_switch_wired_network": true,
+    "min_link_rate_gbps": 2.5,
+    "recommended_link_rate_gbps": 10.0,
+    "require_sentinels": [
+      "coding",
+      "structured_tool_call_json",
+      "multi_step_reasoning"
+    ],
+    "require_openai_compatible_response_fields": [
+      "model",
+      "finish_reason",
+      "usage"
+    ]
+  },
+  "performance": {
+    "min_median_decode_tokens_per_second": 0.5,
+    "max_ttft_seconds_at_4096_prompt": 600,
+    "max_unexplained_stall_seconds": 60,
+    "warmups": 1,
+    "require_per_stage_telemetry": [
+      "compute",
+      "queue",
+      "kv",
+      "seam_bytes",
+      "seam_latency",
+      "rss",
+      "vram",
+      "backend_timing"
+    ],
+    "quality_pass_with_speed_fail_verdict": "stop",
+    "forbid_generalising_results_to_other_hardware": true
+  },
+  "reliability": {
+    "consecutive_clean_cold_starts": 2,
+    "require_cancellation_releases_buffers_and_kv": true,
+    "require_worker_loss_aborts_route": true,
+    "retry_policy": "from_token_zero_on_new_compatible_route",
+    "forbid_silent_kv_migration": true,
+    "require_reject_stale_epoch": true,
+    "require_reject_duplicate_step_id": true,
+    "synthetic_workers_satisfy_alpha": false,
+    "layer_reduced_fixtures_satisfy_alpha": false
+  },
+  "storage": {
+    "mounted_storage_only": true,
+    "forbidden_path_prefixes": [
+      "/home"
+    ],
+    "forbid_secrets_in_logs": true,
+    "forbid_unrestricted_prompt_payloads_in_logs": true
+  },
+  "verdicts": [
+    "alpha",
+    "stop"
+  ],
+  "amendment_policy": "Thresholds are locked before target execution. They may not be weakened, moved, or reinterpreted after results are known. A change requires a new contract_id and contract_version under human review, and the superseded contract is retained.",
+  "contract_sha256": "aab23220280c053a3c14ff559df3cb5c9e1bf7f0f7188c6519e2e9d9ad036ed9"
+}

packages/node/meshnet_node/glm_alpha/data/architecture-snapshot.json

@@ -0,0 +1,91 @@
+{
+  "schema_version": 1,
+  "observed_at": "2026-07-13",
+  "source_repo_id": "zai-org/GLM-5.2",
+  "source_revision": "b4734de4facf877f85769a911abafc5283eab3d9",
+  "source_files": [
+    {
+      "path": "config.json",
+      "size_bytes": 3732,
+      "sha256": "185f93ee6d12548e16a847e279dc0c3c90b1524c970b0866b42fb545747d859a",
+      "url": "https://huggingface.co/zai-org/GLM-5.2/resolve/b4734de4facf877f85769a911abafc5283eab3d9/config.json"
+    },
+    {
+      "path": "chat_template.jinja",
+      "size_bytes": 5076,
+      "sha256": "172dc74a35e1752df75ecfb2b2cf9326d2852bb1379868ebeec9571654489679",
+      "url": "https://huggingface.co/zai-org/GLM-5.2/resolve/b4734de4facf877f85769a911abafc5283eab3d9/chat_template.jinja"
+    },
+    {
+      "path": "generation_config.json",
+      "size_bytes": 194,
+      "sha256": "ac76b43d8683d3b930126870fc8be73d8679308fe752fa1f381096d8354f6a55",
+      "url": "https://huggingface.co/zai-org/GLM-5.2/resolve/b4734de4facf877f85769a911abafc5283eab3d9/generation_config.json"
+    },
+    {
+      "path": "tokenizer_config.json",
+      "size_bytes": 761,
+      "sha256": "98b1271574f41abf89427ae2dda030d94dc9478f0edc5a8bd240db213c6fd5fc",
+      "url": "https://huggingface.co/zai-org/GLM-5.2/resolve/b4734de4facf877f85769a911abafc5283eab3d9/tokenizer_config.json"
+    }
+  ],
+  "architecture": {
+    "architectures": [
+      "GlmMoeDsaForCausalLM"
+    ],
+    "model_type": "glm_moe_dsa",
+    "num_hidden_layers": 78,
+    "num_nextn_predict_layers": 1,
+    "total_artifact_layers": 79,
+    "first_k_dense_replace": 3,
+    "dense_layers": 3,
+    "sparse_moe_layers": 75,
+    "hidden_size": 6144,
+    "intermediate_size": 12288,
+    "moe_intermediate_size": 2048,
+    "n_routed_experts": 256,
+    "num_experts_per_tok": 8,
+    "n_shared_experts": 1,
+    "scoring_func": "sigmoid",
+    "topk_method": "noaux_tc",
+    "norm_topk_prob": true,
+    "routed_scaling_factor": 2.5,
+    "num_attention_heads": 64,
+    "head_dim": 192,
+    "qk_nope_head_dim": 192,
+    "qk_rope_head_dim": 64,
+    "qk_head_dim": 256,
+    "v_head_dim": 256,
+    "kv_lora_rank": 512,
+    "q_lora_rank": 2048,
+    "mla_cached_values_per_token_per_layer": 576,
+    "index_topk": 2048,
+    "index_head_dim": 128,
+    "index_n_heads": 32,
+    "index_topk_freq": 4,
+    "index_skip_topk_offset": 3,
+    "index_share_for_mtp_iteration": true,
+    "indexer_full_layers": 21,
+    "indexer_shared_layers": 57,
+    "indexer_types_sha256": "ec3b4927af83cf02baf37fb10454c40176ec8bf501ae89334b27a9df5fa17025",
+    "max_position_embeddings": 1048576,
+    "vocab_size": 154880,
+    "rope_theta": 8000000,
+    "dtype": "bfloat16",
+    "tie_word_embeddings": false
+  },
+  "reasoning_effort": {
+    "alpha_mode": "max",
+    "rendered_marker": "<|system|>Reasoning Effort: Max",
+    "template_rule": "effective_reasoning_effort = 'high' if reasoning_effort == 'high' else 'max'",
+    "default_is_max": true,
+    "suppressed_when": "enable_thinking is defined and false",
+    "note": "The template recognises exactly one non-max level ('high'); every other value, including an absent reasoning_effort, renders Max. Alpha therefore asserts the rendered 'Reasoning Effort: Max' marker, not merely the presence of a request field."
+  },
+  "notes": [
+    "indexer_types has 78 entries: layers 0-2 are 'full', then a repeating [shared, shared, shared, full] pattern, giving 21 Full producer layers and 57 Shared consumer layers.",
+    "MLA caches kv_lora_rank (512) + qk_rope_head_dim (64) = 576 values per token per backbone layer.",
+    "num_nextn_predict_layers=1 is the NextN/MTP layer present in the artifact. Alpha does not run MTP; the tensors are loaded or explicitly excluded by a certified recipe and must never be silently reinterpreted as a 79th backbone layer.",
+    "Values are derived from the pinned config.json. The runtime must re-derive them from the artifact and fail closed on contradictory metadata; marketing names are not compatibility identity."
+  ]
+}

packages/node/meshnet_node/glm_alpha/data/target-manifest.json

@@ -0,0 +1,90 @@
+{
+  "schema_version": 1,
+  "manifest_version": 1,
+  "observed_at": "2026-07-13",
+  "observed_by": "DGR-017",
+  "alpha_quantization": "UD-IQ1_S",
+  "source_model": {
+    "repo_id": "zai-org/GLM-5.2",
+    "revision": "b4734de4facf877f85769a911abafc5283eab3d9",
+    "last_modified": "2026-07-02T08:08:14.000Z",
+    "weight_license": "mit",
+    "code_documentation_license": "apache-2.0",
+    "url": "https://huggingface.co/zai-org/GLM-5.2",
+    "revision_url": "https://huggingface.co/zai-org/GLM-5.2/tree/b4734de4facf877f85769a911abafc5283eab3d9",
+    "api_url": "https://huggingface.co/api/models/zai-org/GLM-5.2"
+  },
+  "gguf_artifact": {
+    "repo_id": "unsloth/GLM-5.2-GGUF",
+    "revision": "abc55e72527792c6e77069c99b4cb7de16fa9f23",
+    "last_modified": "2026-06-23T15:18:23.000Z",
+    "license": "mit",
+    "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF",
+    "revision_url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/tree/abc55e72527792c6e77069c99b4cb7de16fa9f23",
+    "api_url": "https://huggingface.co/api/models/unsloth/GLM-5.2-GGUF",
+    "quantization": "UD-IQ1_S",
+    "shard_count": 6,
+    "total_bytes": 216715360960,
+    "total_gib": 201.832,
+    "total_gb": 216.715,
+    "shards": [
+      {
+        "index": 1,
+        "path": "UD-IQ1_S/GLM-5.2-UD-IQ1_S-00001-of-00006.gguf",
+        "size_bytes": 9423744,
+        "sha256": "46b6148389219ae45167cb8124fbb18ef7d432daf619b4faf9e06ea80d3f4777",
+        "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00001-of-00006.gguf"
+      },
+      {
+        "index": 2,
+        "path": "UD-IQ1_S/GLM-5.2-UD-IQ1_S-00002-of-00006.gguf",
+        "size_bytes": 49208128256,
+        "sha256": "f2180207285e04fcaa5b8c53ba6e77ad5cc58666b6e7c6b04a5eded3fe8bef09",
+        "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00002-of-00006.gguf"
+      },
+      {
+        "index": 3,
+        "path": "UD-IQ1_S/GLM-5.2-UD-IQ1_S-00003-of-00006.gguf",
+        "size_bytes": 49684417024,
+        "sha256": "b1c0c5a302cc8d5d9ea0bcd4467c01db72c26839f820f7e882079582ea0a8d2b",
+        "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00003-of-00006.gguf"
+      },
+      {
+        "index": 4,
+        "path": "UD-IQ1_S/GLM-5.2-UD-IQ1_S-00004-of-00006.gguf",
+        "size_bytes": 49396052864,
+        "sha256": "a6a42da6975e29f89866dcde2956e9e50e6ea26635fb5063b74f3973f4f863b6",
+        "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00004-of-00006.gguf"
+      },
+      {
+        "index": 5,
+        "path": "UD-IQ1_S/GLM-5.2-UD-IQ1_S-00005-of-00006.gguf",
+        "size_bytes": 49246275936,
+        "sha256": "a4a9851a50db533f21ef824e5d8038f04e6782e7d602d18e5fdd6643f68ccccb",
+        "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00005-of-00006.gguf"
+      },
+      {
+        "index": 6,
+        "path": "UD-IQ1_S/GLM-5.2-UD-IQ1_S-00006-of-00006.gguf",
+        "size_bytes": 19171063136,
+        "sha256": "3b767f55df64e0432d52fcf1a14eb47a1ef3bbc91339e2ae220f38602237d7d7",
+        "url": "https://huggingface.co/unsloth/GLM-5.2-GGUF/resolve/abc55e72527792c6e77069c99b4cb7de16fa9f23/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00006-of-00006.gguf"
+      }
+    ]
+  },
+  "diagnostic_fallback": {
+    "quantization": "UD-IQ1_M",
+    "shard_count": 6,
+    "total_bytes": 228492966624,
+    "total_gib": 212.801,
+    "total_gb": 228.493,
+    "policy": "First diagnostic fallback only if UD-IQ1_S exposes a runtime or quality defect. It does not satisfy the alpha 'lowest published quantization' target unless human review changes the target contract."
+  },
+  "storage": {
+    "mounted_storage_only": true,
+    "forbidden_path_prefixes": [
+      "/home"
+    ],
+    "note": "Model artifacts resolve through the machine-specific .env.<hostname> mounted-drive configuration. A path under /home fails admission closed."
+  }
+}

packages/node/meshnet_node/glm_alpha/manifest.py

@@ -0,0 +1,490 @@
+"""The pinned GLM-5.2 Max target manifest and architecture snapshot.
+
+This module is the *identity* half of the alpha target contract. It answers one
+question: is the artifact on this disk the exact artifact alpha was locked
+against?
+
+Identity is pinned by repository revision **and** by every shard's LFS SHA-256.
+A revision alone is not enough — a repository can be force-pushed, and a tag can
+be moved — and a size alone is not enough, because two different quantizations of
+the same model land within a rounding error of each other. The manifest therefore
+carries both, plus the aggregate byte total, and cross-checks the aggregate
+against the sum of the shards. A manifest whose declared total disagrees with its
+own shards is rejected rather than trusted, because that is the exact shape a
+hand-edited "it fits now" manifest takes.
+
+Nothing here downloads a weight payload. Sizes and hashes come from the Hugging
+Face metadata API (see ``scripts/refresh_glm_target_manifest.py``); verification
+against a local file is DGR-018's job, using the digests locked here.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import json
+import re
+from dataclasses import dataclass
+from importlib.resources import files
+from pathlib import Path
+from typing import Any, Mapping
+
+TARGET_MANIFEST_SCHEMA_VERSION = 1
+ARCHITECTURE_SNAPSHOT_SCHEMA_VERSION = 1
+
+ALPHA_QUANTIZATION = "UD-IQ1_S"
+ALPHA_SHARD_COUNT = 6
+
+_SHA256_RE = re.compile(r"\A[0-9a-f]{64}\Z")
+
+_MANIFEST_RESOURCE = "target-manifest.json"
+_ARCHITECTURE_RESOURCE = "architecture-snapshot.json"
+
+GIB = 1024**3
+GB = 1000**3
+
+
+class GlmTargetError(ValueError):
+    """Raised when the target manifest or architecture snapshot is not the pinned target."""
+
+
+def canonical_sha256(value: Any) -> str:
+    """SHA-256 over canonical JSON — the repository's digest convention."""
+    payload = json.dumps(value, sort_keys=True, separators=(",", ":"), ensure_ascii=False)
+    return hashlib.sha256(payload.encode("utf-8")).hexdigest()
+
+
+def _require_mapping(value: Any, what: str) -> Mapping[str, Any]:
+    if not isinstance(value, Mapping):
+        raise GlmTargetError(f"{what} must be a JSON object, got {type(value).__name__}")
+    return value
+
+
+def _require_text(value: Any, what: str) -> str:
+    if not isinstance(value, str) or not value.strip():
+        raise GlmTargetError(f"{what} must be a non-empty string")
+    return value
+
+
+def _require_int(value: Any, what: str) -> int:
+    if not isinstance(value, int) or isinstance(value, bool):
+        raise GlmTargetError(f"{what} must be an integer, got {type(value).__name__}")
+    return value
+
+
+def _require_sha256(value: Any, what: str) -> str:
+    text = _require_text(value, what)
+    if not _SHA256_RE.match(text):
+        raise GlmTargetError(
+            f"{what} must be a lowercase 64-character hex SHA-256, got {text!r}"
+        )
+    return text
+
+
+def _require_revision(value: Any, what: str) -> str:
+    text = _require_text(value, what)
+    if not re.fullmatch(r"[0-9a-f]{40}", text):
+        raise GlmTargetError(
+            f"{what} must be a full 40-character commit revision, got {text!r}; "
+            "a branch name or short SHA is not an immutable pin"
+        )
+    return text
+
+
+@dataclass(frozen=True)
+class Shard:
+    """One GGUF shard of the alpha artifact."""
+
+    index: int
+    path: str
+    size_bytes: int
+    sha256: str
+    url: str
+
+    def to_dict(self) -> dict:
+        return {
+            "index": self.index,
+            "path": self.path,
+            "size_bytes": self.size_bytes,
+            "sha256": self.sha256,
+            "url": self.url,
+        }
+
+
+@dataclass(frozen=True)
+class TargetManifest:
+    """The pinned, self-consistent GLM-5.2 ``UD-IQ1_S`` target."""
+
+    schema_version: int
+    manifest_version: int
+    observed_at: str
+    quantization: str
+    source_repo_id: str
+    source_revision: str
+    source_license: str
+    gguf_repo_id: str
+    gguf_revision: str
+    gguf_license: str
+    total_bytes: int
+    shards: tuple[Shard, ...]
+    raw: Mapping[str, Any]
+    source: str = "<memory>"
+
+    @property
+    def total_gib(self) -> float:
+        return self.total_bytes / GIB
+
+    @property
+    def total_gb(self) -> float:
+        return self.total_bytes / GB
+
+    def shard(self, index: int) -> Shard:
+        for shard in self.shards:
+            if shard.index == index:
+                return shard
+        raise GlmTargetError(f"shard {index} is not in {self.source}")
+
+    @property
+    def digest(self) -> str:
+        """Stable identity of this manifest, for the DGR-003 runtime recipe."""
+        return canonical_sha256(self.raw)
+
+    def to_dict(self) -> dict:
+        return dict(self.raw)
+
+
+def _parse_shards(raw: Any, expected_count: int, expected_total: int) -> tuple[Shard, ...]:
+    if not isinstance(raw, list):
+        raise GlmTargetError("gguf_artifact.shards must be a JSON array")
+    if len(raw) != expected_count:
+        raise GlmTargetError(
+            f"the alpha artifact has exactly {expected_count} shards, "
+            f"but the manifest lists {len(raw)}"
+        )
+
+    shards: list[Shard] = []
+    seen_index: set[int] = set()
+    seen_sha: set[str] = set()
+    for position, entry in enumerate(raw):
+        item = _require_mapping(entry, f"shards[{position}]")
+        index = _require_int(item.get("index"), f"shards[{position}].index")
+        if index in seen_index:
+            raise GlmTargetError(f"duplicate shard index {index} in the manifest")
+        seen_index.add(index)
+        size_bytes = _require_int(item.get("size_bytes"), f"shards[{index}].size_bytes")
+        if size_bytes <= 0:
+            raise GlmTargetError(f"shards[{index}].size_bytes must be positive")
+        sha256 = _require_sha256(item.get("sha256"), f"shards[{index}].sha256")
+        if sha256 in seen_sha:
+            raise GlmTargetError(
+                f"shard {index} repeats SHA-256 {sha256}; two distinct shards cannot "
+                "have the same content digest"
+            )
+        seen_sha.add(sha256)
+        shards.append(
+            Shard(
+                index=index,
+                path=_require_text(item.get("path"), f"shards[{index}].path"),
+                size_bytes=size_bytes,
+                sha256=sha256,
+                url=_require_text(item.get("url"), f"shards[{index}].url"),
+            )
+        )
+
+    expected_indices = set(range(1, expected_count + 1))
+    if seen_index != expected_indices:
+        missing = sorted(expected_indices - seen_index)
+        raise GlmTargetError(
+            f"the manifest is missing shard(s) {missing}; all {expected_count} "
+            "shards of the alpha artifact must be pinned"
+        )
+
+    summed = sum(shard.size_bytes for shard in shards)
+    if summed != expected_total:
+        raise GlmTargetError(
+            f"declared total_bytes {expected_total} does not equal the sum of the "
+            f"shard sizes {summed}; the manifest is not self-consistent"
+        )
+
+    return tuple(sorted(shards, key=lambda shard: shard.index))
+
+
+def parse_target_manifest(data: Any, source: str = "<memory>") -> TargetManifest:
+    """Validate an already-decoded target manifest, failing closed."""
+    doc = _require_mapping(data, f"manifest root in {source}")
+
+    schema_version = _require_int(doc.get("schema_version"), f"'schema_version' in {source}")
+    if schema_version != TARGET_MANIFEST_SCHEMA_VERSION:
+        raise GlmTargetError(
+            f"{source} declares target-manifest schema version {schema_version}, "
+            f"but this node reads version {TARGET_MANIFEST_SCHEMA_VERSION}"
+        )
+
+    quantization = _require_text(doc.get("alpha_quantization"), f"'alpha_quantization' in {source}")
+    if quantization != ALPHA_QUANTIZATION:
+        raise GlmTargetError(
+            f"{source} pins quantization {quantization!r}, but the locked alpha "
+            f"quantization is {ALPHA_QUANTIZATION!r}; a different quantization is a "
+            "different target and requires a human contract change"
+        )
+
+    source_model = _require_mapping(doc.get("source_model"), f"'source_model' in {source}")
+    gguf = _require_mapping(doc.get("gguf_artifact"), f"'gguf_artifact' in {source}")
+
+    gguf_quant = _require_text(gguf.get("quantization"), f"gguf_artifact.quantization in {source}")
+    if gguf_quant != quantization:
+        raise GlmTargetError(
+            f"{source} declares alpha_quantization {quantization!r} but the GGUF "
+            f"artifact block says {gguf_quant!r}"
+        )
+
+    shard_count = _require_int(gguf.get("shard_count"), f"gguf_artifact.shard_count in {source}")
+    if shard_count != ALPHA_SHARD_COUNT:
+        raise GlmTargetError(
+            f"{source} declares {shard_count} shards; the pinned alpha artifact has "
+            f"exactly {ALPHA_SHARD_COUNT}"
+        )
+
+    total_bytes = _require_int(gguf.get("total_bytes"), f"gguf_artifact.total_bytes in {source}")
+    shards = _parse_shards(gguf.get("shards"), shard_count, total_bytes)
+
+    return TargetManifest(
+        schema_version=schema_version,
+        manifest_version=_require_int(doc.get("manifest_version"), f"'manifest_version' in {source}"),
+        observed_at=_require_text(doc.get("observed_at"), f"'observed_at' in {source}"),
+        quantization=quantization,
+        source_repo_id=_require_text(source_model.get("repo_id"), "source_model.repo_id"),
+        source_revision=_require_revision(source_model.get("revision"), "source_model.revision"),
+        source_license=_require_text(source_model.get("weight_license"), "source_model.weight_license"),
+        gguf_repo_id=_require_text(gguf.get("repo_id"), "gguf_artifact.repo_id"),
+        gguf_revision=_require_revision(gguf.get("revision"), "gguf_artifact.revision"),
+        gguf_license=_require_text(gguf.get("license"), "gguf_artifact.license"),
+        total_bytes=total_bytes,
+        shards=shards,
+        raw=doc,
+        source=source,
+    )
+
+
+@dataclass(frozen=True)
+class ArchitectureSnapshot:
+    """Architecture-critical metadata derived from the pinned ``config.json``."""
+
+    schema_version: int
+    source_repo_id: str
+    source_revision: str
+    architecture: Mapping[str, Any]
+    reasoning_effort: Mapping[str, Any]
+    source_files: Mapping[str, Mapping[str, Any]]
+    raw: Mapping[str, Any]
+    source: str = "<memory>"
+
+    def __getitem__(self, key: str) -> Any:
+        if key not in self.architecture:
+            raise GlmTargetError(f"architecture field {key!r} is missing from {self.source}")
+        return self.architecture[key]
+
+    @property
+    def digest(self) -> str:
+        return canonical_sha256(self.raw)
+
+    def file_sha256(self, path: str) -> str:
+        entry = self.source_files.get(path)
+        if entry is None:
+            raise GlmTargetError(f"{path!r} is not pinned in {self.source}")
+        return str(entry["sha256"])
+
+    def to_dict(self) -> dict:
+        return dict(self.raw)
+
+
+# Fields the distributed runtime cannot plan or shard without. Absent or
+# contradictory values fail closed rather than defaulting.
+REQUIRED_ARCHITECTURE_FIELDS: tuple[str, ...] = (
+    "model_type",
+    "num_hidden_layers",
+    "num_nextn_predict_layers",
+    "total_artifact_layers",
+    "first_k_dense_replace",
+    "dense_layers",
+    "sparse_moe_layers",
+    "hidden_size",
+    "n_routed_experts",
+    "num_experts_per_tok",
+    "n_shared_experts",
+    "kv_lora_rank",
+    "qk_rope_head_dim",
+    "mla_cached_values_per_token_per_layer",
+    "index_topk",
+    "index_head_dim",
+    "indexer_full_layers",
+    "indexer_shared_layers",
+    "max_position_embeddings",
+    "vocab_size",
+)
+
+
+def parse_architecture_snapshot(data: Any, source: str = "<memory>") -> ArchitectureSnapshot:
+    """Validate an architecture snapshot and its internal arithmetic."""
+    doc = _require_mapping(data, f"snapshot root in {source}")
+
+    schema_version = _require_int(doc.get("schema_version"), f"'schema_version' in {source}")
+    if schema_version != ARCHITECTURE_SNAPSHOT_SCHEMA_VERSION:
+        raise GlmTargetError(
+            f"{source} declares architecture-snapshot schema version {schema_version}, "
+            f"but this node reads version {ARCHITECTURE_SNAPSHOT_SCHEMA_VERSION}"
+        )
+
+    arch = _require_mapping(doc.get("architecture"), f"'architecture' in {source}")
+    missing = [field for field in REQUIRED_ARCHITECTURE_FIELDS if field not in arch]
+    if missing:
+        raise GlmTargetError(
+            f"{source} is missing architecture-critical field(s) {missing}; the "
+            "runtime cannot shard or plan an architecture it cannot fully describe"
+        )
+
+    layers = _require_int(arch["num_hidden_layers"], "num_hidden_layers")
+    nextn = _require_int(arch["num_nextn_predict_layers"], "num_nextn_predict_layers")
+    total_layers = _require_int(arch["total_artifact_layers"], "total_artifact_layers")
+    if total_layers != layers + nextn:
+        raise GlmTargetError(
+            f"total_artifact_layers {total_layers} != num_hidden_layers {layers} + "
+            f"num_nextn_predict_layers {nextn}; the NextN layer must be counted "
+            "explicitly, never folded into the backbone"
+        )
+
+    dense = _require_int(arch["dense_layers"], "dense_layers")
+    sparse = _require_int(arch["sparse_moe_layers"], "sparse_moe_layers")
+    if dense != _require_int(arch["first_k_dense_replace"], "first_k_dense_replace"):
+        raise GlmTargetError("dense_layers must equal first_k_dense_replace")
+    if dense + sparse != layers:
+        raise GlmTargetError(
+            f"dense_layers {dense} + sparse_moe_layers {sparse} != num_hidden_layers {layers}"
+        )
+
+    full = _require_int(arch["indexer_full_layers"], "indexer_full_layers")
+    shared = _require_int(arch["indexer_shared_layers"], "indexer_shared_layers")
+    if full + shared != layers:
+        raise GlmTargetError(
+            f"indexer_full_layers {full} + indexer_shared_layers {shared} != "
+            f"num_hidden_layers {layers}; every layer holds exactly one IndexShare role"
+        )
+    if full <= 0:
+        raise GlmTargetError(
+            "indexer_full_layers must be positive; a route with no Full producer layer "
+            "has no index for its Shared consumers to reuse"
+        )
+
+    mla = _require_int(
+        arch["mla_cached_values_per_token_per_layer"], "mla_cached_values_per_token_per_layer"
+    )
+    expected_mla = _require_int(arch["kv_lora_rank"], "kv_lora_rank") + _require_int(
+        arch["qk_rope_head_dim"], "qk_rope_head_dim"
+    )
+    if mla != expected_mla:
+        raise GlmTargetError(
+            f"mla_cached_values_per_token_per_layer {mla} != kv_lora_rank + "
+            f"qk_rope_head_dim ({expected_mla})"
+        )
+
+    reasoning = _require_mapping(doc.get("reasoning_effort"), f"'reasoning_effort' in {source}")
+    if reasoning.get("alpha_mode") != "max":
+        raise GlmTargetError(
+            f"{source} does not lock reasoning_effort=max; alpha is defined as the "
+            "Max reasoning mode of this exact checkpoint"
+        )
+    _require_text(reasoning.get("rendered_marker"), "reasoning_effort.rendered_marker")
+
+    files_raw = doc.get("source_files")
+    if not isinstance(files_raw, list) or not files_raw:
+        raise GlmTargetError(f"'source_files' in {source} must be a non-empty JSON array")
+    source_files: dict[str, Mapping[str, Any]] = {}
+    for position, entry in enumerate(files_raw):
+        item = _require_mapping(entry, f"source_files[{position}]")
+        path = _require_text(item.get("path"), f"source_files[{position}].path")
+        _require_sha256(item.get("sha256"), f"source_files[{path}].sha256")
+        source_files[path] = item
+    for required in ("config.json", "chat_template.jinja"):
+        if required not in source_files:
+            raise GlmTargetError(
+                f"{source} does not pin {required!r}; config and chat-template drift "
+                "silently changes runtime semantics"
+            )
+
+    return ArchitectureSnapshot(
+        schema_version=schema_version,
+        source_repo_id=_require_text(doc.get("source_repo_id"), "source_repo_id"),
+        source_revision=_require_revision(doc.get("source_revision"), "source_revision"),
+        architecture=arch,
+        reasoning_effort=reasoning,
+        source_files=source_files,
+        raw=doc,
+        source=source,
+    )
+
+
+def _read_resource(resource: str, path: Path | None) -> tuple[str, str]:
+    if path is not None:
+        try:
+            return str(path), path.read_text(encoding="utf-8")
+        except OSError as exc:
+            raise GlmTargetError(f"cannot read {path}: {exc.strerror or exc}") from exc
+    source = f"packaged {resource}"
+    try:
+        raw = files("meshnet_node.glm_alpha").joinpath("data", resource).read_text(encoding="utf-8")
+    except (OSError, FileNotFoundError, ModuleNotFoundError) as exc:
+        raise GlmTargetError(
+            f"{source} is missing from this node installation ({type(exc).__name__})"
+        ) from exc
+    return source, raw
+
+
+def _load_json(resource: str, path: Path | None) -> tuple[str, Any]:
+    source, raw = _read_resource(resource, path)
+    try:
+        return source, json.loads(raw)
+    except json.JSONDecodeError as exc:
+        raise GlmTargetError(
+            f"{source} is not valid JSON: {exc.msg} at line {exc.lineno} column {exc.colno}"
+        ) from exc
+
+
+def load_target_manifest(path: Path | None = None) -> TargetManifest:
+    """Load the packaged target manifest, or one at ``path``."""
+    source, data = _load_json(_MANIFEST_RESOURCE, path)
+    return parse_target_manifest(data, source=source)
+
+
+def load_architecture_snapshot(path: Path | None = None) -> ArchitectureSnapshot:
+    """Load the packaged architecture snapshot, or one at ``path``."""
+    source, data = _load_json(_ARCHITECTURE_RESOURCE, path)
+    return parse_architecture_snapshot(data, source=source)
+
+
+def require_pinned_target(
+    manifest: TargetManifest,
+    snapshot: ArchitectureSnapshot,
+    *,
+    expected_source_revision: str,
+    expected_gguf_revision: str,
+) -> None:
+    """Reject any target whose revisions are not the ones alpha was locked against.
+
+    Callers pass the revisions from the locked alpha contract, so a swapped
+    manifest cannot quietly re-point the target at a different upstream commit.
+    """
+    if manifest.source_revision != expected_source_revision:
+        raise GlmTargetError(
+            f"source revision {manifest.source_revision} does not match the locked "
+            f"alpha revision {expected_source_revision}"
+        )
+    if manifest.gguf_revision != expected_gguf_revision:
+        raise GlmTargetError(
+            f"GGUF revision {manifest.gguf_revision} does not match the locked alpha "
+            f"revision {expected_gguf_revision}"
+        )
+    if snapshot.source_revision != manifest.source_revision:
+        raise GlmTargetError(
+            f"the architecture snapshot was taken at {snapshot.source_revision} but the "
+            f"manifest pins {manifest.source_revision}; config metadata and weights must "
+            "come from one revision"
+        )

packages/node/meshnet_node/glm_alpha/planner.py

@@ -0,0 +1,522 @@
+"""Deterministic memory, KV, and network planner for the GLM-5.2 Max alpha route.
+
+Everything here is arithmetic over the exact pinned artifact bytes and the exact
+pinned architecture. There is no measurement, no probing, and no heuristic tuned
+to a result — the planner is written *before* the target runs so that a later
+story cannot discover a topology that "works" and then rationalise it.
+
+Three ideas do the real work.
+
+**Unified memory is one pool.** On an integrated-GPU machine the "VRAM" the driver
+reports is carved out of the same physical DRAM the OS is already counting. Adding
+them produces a node that appears to hold twice what it holds, and the failure mode
+is not a clean admission rejection — it is an OOM or a swap-thrash halfway through
+a 200 GiB load. :class:`NodeMemory` therefore refuses to be constructed from an
+additive claim about one shared pool.
+
+**The reserve is not optional headroom.** Weights plus KV are not the whole
+resident cost: backend workspaces, quantization scratch, the graph plan, the
+process, and the OS all live outside them, and the largest of those scale with the
+backend rather than with the shard. Alpha reserves ``max(20% of physically usable
+memory, 8 GiB)`` per node, and the *remainder* is the placement budget.
+
+**Equal layer counts are not equal bytes.** Embeddings and the output head are
+endpoint-only; three layers are dense and 75 are MoE; shared experts, indexer
+tensors, and quant block alignment all skew the per-node share. Until DGR-018/019
+report measured per-tensor placement, the planner carries an explicit
+:data:`PLACEMENT_IMBALANCE_FACTOR` and reports the arithmetic minimum and the
+recommended count as two separate numbers. The arithmetic minimum is a fit probe;
+it is admissible only with exact measured placement evidence behind it.
+"""
+
+from __future__ import annotations
+
+import math
+from dataclasses import dataclass
+from typing import Literal
+
+from .manifest import GIB, ArchitectureSnapshot, TargetManifest
+
+# Q8_0 stores 32 int8 quants plus one fp16 scale per block: 34 bytes / 32 values.
+Q8_0_BYTES_PER_VALUE = 34 / 32
+F16_BYTES_PER_VALUE = 2.0
+
+KV_DTYPES: dict[str, float] = {
+    "Q8_0": Q8_0_BYTES_PER_VALUE,
+    "F16": F16_BYTES_PER_VALUE,
+}
+
+# The alpha KV configuration, locked by the roadmap.
+ALPHA_KV_DTYPE = "Q8_0"
+ALPHA_CONTEXT_TOKENS = 16384
+ALPHA_CONCURRENCY = 1
+
+# The reserve every node holds outside its weight-plus-KV placement budget.
+RESERVE_FRACTION = 0.20
+RESERVE_FLOOR_GIB = 8.0
+
+# The aggregate runtime-accessible memory at which the artifact *just* fits.
+# This is an experimental hard-fit floor, not an operational envelope: it has no
+# room for a backend that allocates more scratch than another, and none for the
+# imbalance below.
+AGGREGATE_HARD_FIT_FLOOR_GIB = 224.0
+
+# How much more than an equal share the worst-placed node is expected to hold.
+# 1.10 is the roadmap's recommended-topology column expressed as arithmetic: it
+# reproduces 10 / 6 / 5 / 3 / 3 nodes for the 32 / 48 / 64 / 96 / 128 GiB tiers.
+# DGR-019 must replace it with measured per-tensor placement.
+PLACEMENT_IMBALANCE_FACTOR = 1.10
+
+# Alpha network floor. A link rate is a bandwidth claim, never a speed claim.
+MIN_LINK_RATE_GBPS = 2.5
+RECOMMENDED_LINK_RATE_GBPS = 10.0
+
+BF16_BYTES = 2
+DSA_SIDEBAND_INT32_BYTES = 4
+
+IndexerLayout = Literal["optimized", "conservative"]
+
+
+class ResourcePlanError(ValueError):
+    """Raised when a node or route cannot be accounted for honestly."""
+
+
+@dataclass(frozen=True)
+class NodeMemory:
+    """One node's physically usable memory, counted once.
+
+    ``physical_usable_gib`` is what the node can actually place bytes into after
+    firmware and fixed carve-outs — not the marketing capacity, and not a sum of
+    two views of the same DRAM.
+    """
+
+    name: str
+    physical_usable_gib: float
+    unified: bool
+
+    def __post_init__(self) -> None:
+        if not isinstance(self.name, str) or not self.name.strip():
+            raise ResourcePlanError("node name must be a non-empty physical-host identity")
+        if (
+            isinstance(self.physical_usable_gib, bool)
+            or not isinstance(self.physical_usable_gib, (int, float))
+            or not math.isfinite(self.physical_usable_gib)
+            or self.physical_usable_gib <= 0
+        ):
+            raise ResourcePlanError(
+                f"node {self.name!r} must declare finite positive usable memory"
+            )
+
+    @classmethod
+    def from_host(
+        cls,
+        name: str,
+        *,
+        system_ram_gib: float,
+        gpu_memory_gib: float = 0.0,
+        unified: bool,
+    ) -> "NodeMemory":
+        """Build a node from a host's reported RAM and GPU memory.
+
+        On a unified machine the GPU memory *is* system RAM, so it is counted once
+        and never added. Passing a non-zero ``gpu_memory_gib`` alongside
+        ``unified=True`` is the double-count this project has already decided is a
+        bug (RALPH-CONTEXT runtime decision 16), so it is rejected rather than
+        silently discarded: a caller who believes an integrated GPU adds memory has
+        a wrong model of the machine, and quietly ignoring the argument would let
+        that belief survive.
+        """
+        if not isinstance(unified, bool):
+            raise ResourcePlanError(f"node {name!r} unified flag must be boolean")
+        for value, label, allow_zero in (
+            (system_ram_gib, "system RAM", False),
+            (gpu_memory_gib, "GPU memory", True),
+        ):
+            if (
+                isinstance(value, bool)
+                or not isinstance(value, (int, float))
+                or not math.isfinite(value)
+                or value < 0
+                or (not allow_zero and value == 0)
+            ):
+                qualifier = "finite non-negative" if allow_zero else "finite positive"
+                raise ResourcePlanError(f"node {name!r} must declare {qualifier} {label}")
+        if unified:
+            if gpu_memory_gib:
+                raise ResourcePlanError(
+                    f"node {name!r} declares unified memory and {gpu_memory_gib} GiB of "
+                    "separate GPU memory. Integrated-GPU memory is carved out of the same "
+                    "physical DRAM as system RAM; adding them double-counts one pool. "
+                    "Pass unified=True with system_ram_gib only."
+                )
+            usable = system_ram_gib
+        else:
+            usable = system_ram_gib + gpu_memory_gib
+        return cls(name=name, physical_usable_gib=usable, unified=unified)
+
+    @property
+    def reserve_gib(self) -> float:
+        """``max(20% of physically usable memory, 8 GiB)``."""
+        return max(RESERVE_FRACTION * self.physical_usable_gib, RESERVE_FLOOR_GIB)
+
+    @property
+    def placement_budget_gib(self) -> float:
+        """What remains for weights plus KV after the reserve."""
+        return self.physical_usable_gib - self.reserve_gib
+
+
+def kv_bytes(
+    snapshot: ArchitectureSnapshot,
+    *,
+    context_tokens: int = ALPHA_CONTEXT_TOKENS,
+    concurrency: int = ALPHA_CONCURRENCY,
+    dtype: str = ALPHA_KV_DTYPE,
+    indexer_layout: IndexerLayout = "conservative",
+    include_indexer: bool = True,
+) -> int:
+    """Bytes of MLA (and DSA indexer) KV cache for the whole model.
+
+    ``indexer_layout`` is the honest part. Correct DSA only needs indexer keys for
+    the Full producer layers, but the current experimental implementation may
+    allocate them across every backbone layer. Alpha budgets ``conservative``
+    (all 78) so that a route admitted by this planner cannot be surprised by the
+    implementation it actually gets.
+    """
+    if (
+        not isinstance(context_tokens, int)
+        or isinstance(context_tokens, bool)
+        or context_tokens <= 0
+        or not isinstance(concurrency, int)
+        or isinstance(concurrency, bool)
+        or concurrency <= 0
+    ):
+        raise ResourcePlanError("context_tokens and concurrency must be positive integers")
+    if dtype not in KV_DTYPES:
+        raise ResourcePlanError(
+            f"unsupported KV dtype {dtype!r}; alpha locks {ALPHA_KV_DTYPE} "
+            f"(known: {', '.join(sorted(KV_DTYPES))})"
+        )
+    bytes_per_value = KV_DTYPES[dtype]
+
+    layers = int(snapshot["num_hidden_layers"])
+    mla_values = int(snapshot["mla_cached_values_per_token_per_layer"])
+    total_values = mla_values * layers
+
+    if include_indexer:
+        if indexer_layout == "optimized":
+            indexer_layers = int(snapshot["indexer_full_layers"])
+        elif indexer_layout == "conservative":
+            indexer_layers = layers
+        else:  # pragma: no cover - Literal keeps this unreachable from typed callers
+            raise ResourcePlanError(f"unknown indexer_layout {indexer_layout!r}")
+        total_values += int(snapshot["index_head_dim"]) * indexer_layers
+
+    return int(total_values * context_tokens * concurrency * bytes_per_value)
+
+
+@dataclass(frozen=True)
+class TopologyPlan:
+    """The node count a homogeneous tier needs, and how it was reached."""
+
+    physical_usable_gib: float
+    reserve_gib: float
+    placement_budget_gib: float
+    weight_gib: float
+    kv_gib: float
+    total_placement_gib: float
+    arithmetic_minimum_nodes: int
+    recommended_nodes: int
+    imbalance_factor: float
+
+    @property
+    def is_arithmetic_minimum_topology(self) -> bool:
+        """True when the recommendation offers no imbalance headroom at all."""
+        return self.recommended_nodes == self.arithmetic_minimum_nodes
+
+    def to_dict(self) -> dict:
+        return {
+            "physical_usable_gib": round(self.physical_usable_gib, 3),
+            "reserve_gib": round(self.reserve_gib, 3),
+            "placement_budget_gib": round(self.placement_budget_gib, 3),
+            "weight_gib": round(self.weight_gib, 3),
+            "kv_gib": round(self.kv_gib, 3),
+            "total_placement_gib": round(self.total_placement_gib, 3),
+            "arithmetic_minimum_nodes": self.arithmetic_minimum_nodes,
+            "recommended_nodes": self.recommended_nodes,
+            "imbalance_factor": self.imbalance_factor,
+        }
+
+
+def plan_topology(
+    manifest: TargetManifest,
+    snapshot: ArchitectureSnapshot,
+    *,
+    physical_usable_gib: float,
+    context_tokens: int = ALPHA_CONTEXT_TOKENS,
+    concurrency: int = ALPHA_CONCURRENCY,
+    kv_dtype: str = ALPHA_KV_DTYPE,
+    indexer_layout: IndexerLayout = "conservative",
+    imbalance_factor: float = PLACEMENT_IMBALANCE_FACTOR,
+) -> TopologyPlan:
+    """Minimum and recommended node count for a homogeneous tier of this size."""
+    if (
+        isinstance(imbalance_factor, bool)
+        or not isinstance(imbalance_factor, (int, float))
+        or not math.isfinite(imbalance_factor)
+        or imbalance_factor < 1.0
+    ):
+        raise ResourcePlanError(
+            "imbalance_factor must be finite and at least 1.0; a lower value would "
+            "assume the worst-placed node holds less than an equal share"
+        )
+
+    node = NodeMemory(
+        name=f"{physical_usable_gib:g}GiB-tier",
+        physical_usable_gib=physical_usable_gib,
+        unified=False,
+    )
+    budget = node.placement_budget_gib
+    if budget <= 0:
+        raise ResourcePlanError(
+            f"a {physical_usable_gib:g} GiB node has no placement budget after its "
+            f"{node.reserve_gib:.1f} GiB reserve"
+        )
+
+    weight_gib = manifest.total_bytes / GIB
+    kv_gib = (
+        kv_bytes(
+            snapshot,
+            context_tokens=context_tokens,
+            concurrency=concurrency,
+            dtype=kv_dtype,
+            indexer_layout=indexer_layout,
+        )
+        / GIB
+    )
+    total = weight_gib + kv_gib
+
+    return TopologyPlan(
+        physical_usable_gib=physical_usable_gib,
+        reserve_gib=node.reserve_gib,
+        placement_budget_gib=budget,
+        weight_gib=weight_gib,
+        kv_gib=kv_gib,
+        total_placement_gib=total,
+        arithmetic_minimum_nodes=math.ceil(total / budget),
+        recommended_nodes=math.ceil(total * imbalance_factor / budget),
+        imbalance_factor=imbalance_factor,
+    )
+
+
+@dataclass(frozen=True)
+class RouteFit:
+    """Whether a concrete, possibly heterogeneous set of nodes can hold the target."""
+
+    node_count: int
+    aggregate_usable_gib: float
+    aggregate_placement_budget_gib: float
+    required_placement_gib: float
+    fits: bool
+    meets_hard_fit_floor: bool
+    no_single_node_can_admit_target: bool
+    headroom_gib: float
+    reasons: tuple[str, ...]
+
+    def to_dict(self) -> dict:
+        return {
+            "node_count": self.node_count,
+            "aggregate_usable_gib": round(self.aggregate_usable_gib, 3),
+            "aggregate_placement_budget_gib": round(self.aggregate_placement_budget_gib, 3),
+            "required_placement_gib": round(self.required_placement_gib, 3),
+            "fits": self.fits,
+            "meets_hard_fit_floor": self.meets_hard_fit_floor,
+            "no_single_node_can_admit_target": self.no_single_node_can_admit_target,
+            "headroom_gib": round(self.headroom_gib, 3),
+            "reasons": list(self.reasons),
+        }
+
+
+def plan_route(
+    manifest: TargetManifest,
+    snapshot: ArchitectureSnapshot,
+    nodes: list[NodeMemory],
+    *,
+    context_tokens: int = ALPHA_CONTEXT_TOKENS,
+    concurrency: int = ALPHA_CONCURRENCY,
+    kv_dtype: str = ALPHA_KV_DTYPE,
+    indexer_layout: IndexerLayout = "conservative",
+) -> RouteFit:
+    """Evaluate a concrete route. Every node's memory is already counted once."""
+    if len(nodes) < 2:
+        raise ResourcePlanError(
+            "the alpha target is distributed by definition; a route needs at least two "
+            "physical nodes"
+        )
+    names = [node.name for node in nodes]
+    if len(set(names)) != len(names):
+        raise ResourcePlanError(
+            "duplicate node names in the route; one physical machine counted twice is "
+            "the same double-count as adding integrated-GPU memory to system RAM"
+        )
+
+    weight_gib = manifest.total_bytes / GIB
+    kv_gib = (
+        kv_bytes(
+            snapshot,
+            context_tokens=context_tokens,
+            concurrency=concurrency,
+            dtype=kv_dtype,
+            indexer_layout=indexer_layout,
+        )
+        / GIB
+    )
+    required = weight_gib + kv_gib
+
+    aggregate_usable = sum(node.physical_usable_gib for node in nodes)
+    aggregate_budget = sum(node.placement_budget_gib for node in nodes)
+    fits = aggregate_budget >= required
+    largest_budget = max(node.placement_budget_gib for node in nodes)
+    no_single_node = largest_budget < required
+
+    reasons: list[str] = []
+    if not fits:
+        reasons.append(
+            f"aggregate placement budget {aggregate_budget:.1f} GiB is below the "
+            f"{required:.1f} GiB the target needs after each node's reserve"
+        )
+    if not no_single_node:
+        reasons.append(
+            "at least one node could admit the complete target alone; that is a "
+            "single-host run, not distributed alpha"
+        )
+    if aggregate_usable < AGGREGATE_HARD_FIT_FLOOR_GIB:
+        reasons.append(
+            f"aggregate usable memory {aggregate_usable:.1f} GiB is below the "
+            f"{AGGREGATE_HARD_FIT_FLOOR_GIB:g} GiB experimental hard-fit floor"
+        )
+
+    return RouteFit(
+        node_count=len(nodes),
+        aggregate_usable_gib=aggregate_usable,
+        aggregate_placement_budget_gib=aggregate_budget,
+        required_placement_gib=required,
+        fits=fits,
+        meets_hard_fit_floor=aggregate_usable >= AGGREGATE_HARD_FIT_FLOOR_GIB,
+        no_single_node_can_admit_target=no_single_node,
+        headroom_gib=aggregate_budget - required,
+        reasons=tuple(reasons),
+    )
+
+
+@dataclass(frozen=True)
+class SeamPlan:
+    """Bytes and latency across the activation seams of a route.
+
+    Bandwidth and latency are reported separately on purpose. Decode moves almost
+    nothing — 12 KiB per token per seam — so a faster link barely helps it. What
+    decode pays is *serial*: every generated token crosses every seam in order, so
+    the cost that matters is ``seams x per-hop latency``. A route that claims to be
+    fast because it is on 10 GbE has confused the two.
+    """
+
+    node_count: int
+    seam_count: int
+    hidden_size: int
+    bytes_per_token_per_seam: int
+    prefill_bytes_per_seam: int
+    decode_bytes_per_seam_per_token: int
+    dsa_sideband_bytes_per_query: int
+    link_rate_gbps: float
+    meets_alpha_minimum: bool
+    is_recommended_link: bool
+    decode_serialization_ms_per_token: float
+    decode_latency_ms_per_token: float
+    decode_bandwidth_share_ms_per_token: float
+    prefill_serialization_ms: float
+
+    def to_dict(self) -> dict:
+        return {
+            "node_count": self.node_count,
+            "seam_count": self.seam_count,
+            "hidden_size": self.hidden_size,
+            "bytes_per_token_per_seam": self.bytes_per_token_per_seam,
+            "prefill_bytes_per_seam": self.prefill_bytes_per_seam,
+            "decode_bytes_per_seam_per_token": self.decode_bytes_per_seam_per_token,
+            "dsa_sideband_bytes_per_query": self.dsa_sideband_bytes_per_query,
+            "link_rate_gbps": self.link_rate_gbps,
+            "meets_alpha_minimum": self.meets_alpha_minimum,
+            "is_recommended_link": self.is_recommended_link,
+            "decode_serialization_ms_per_token": round(self.decode_serialization_ms_per_token, 4),
+            "decode_latency_ms_per_token": round(self.decode_latency_ms_per_token, 4),
+            "decode_bandwidth_share_ms_per_token": round(
+                self.decode_bandwidth_share_ms_per_token, 4
+            ),
+            "prefill_serialization_ms": round(self.prefill_serialization_ms, 3),
+        }
+
+
+def plan_seams(
+    snapshot: ArchitectureSnapshot,
+    *,
+    node_count: int,
+    context_tokens: int = ALPHA_CONTEXT_TOKENS,
+    link_rate_gbps: float = MIN_LINK_RATE_GBPS,
+    per_hop_latency_ms: float = 0.5,
+) -> SeamPlan:
+    """Model seam bytes, wire serialization, and serial per-hop latency separately."""
+    if not isinstance(node_count, int) or isinstance(node_count, bool) or node_count < 2:
+        raise ResourcePlanError("a seam exists only between two nodes")
+    if not isinstance(context_tokens, int) or isinstance(context_tokens, bool) or context_tokens <= 0:
+        raise ResourcePlanError("context_tokens must be a positive integer")
+    if (
+        isinstance(link_rate_gbps, bool)
+        or not isinstance(link_rate_gbps, (int, float))
+        or not math.isfinite(link_rate_gbps)
+        or link_rate_gbps <= 0
+    ):
+        raise ResourcePlanError("link_rate_gbps must be finite and positive")
+    if (
+        isinstance(per_hop_latency_ms, bool)
+        or not isinstance(per_hop_latency_ms, (int, float))
+        or not math.isfinite(per_hop_latency_ms)
+        or per_hop_latency_ms < 0
+    ):
+        raise ResourcePlanError("per_hop_latency_ms must be finite and non-negative")
+
+    hidden = int(snapshot["hidden_size"])
+    bytes_per_token = hidden * BF16_BYTES
+    seams = node_count - 1
+
+    bits_per_ms = link_rate_gbps * 1e9 / 1e3
+    decode_serialization_ms = (bytes_per_token * 8) / bits_per_ms
+    prefill_serialization_ms = (bytes_per_token * context_tokens * 8) / bits_per_ms
+
+    return SeamPlan(
+        node_count=node_count,
+        seam_count=seams,
+        hidden_size=hidden,
+        bytes_per_token_per_seam=bytes_per_token,
+        prefill_bytes_per_seam=bytes_per_token * context_tokens,
+        decode_bytes_per_seam_per_token=bytes_per_token,
+        dsa_sideband_bytes_per_query=int(snapshot["index_topk"]) * DSA_SIDEBAND_INT32_BYTES,
+        link_rate_gbps=link_rate_gbps,
+        meets_alpha_minimum=link_rate_gbps >= MIN_LINK_RATE_GBPS,
+        is_recommended_link=link_rate_gbps >= RECOMMENDED_LINK_RATE_GBPS,
+        decode_serialization_ms_per_token=decode_serialization_ms * seams,
+        decode_latency_ms_per_token=per_hop_latency_ms * seams,
+        decode_bandwidth_share_ms_per_token=decode_serialization_ms * seams,
+        prefill_serialization_ms=prefill_serialization_ms * seams,
+    )
+
+
+ALPHA_TIERS_GIB: tuple[float, ...] = (32.0, 48.0, 64.0, 96.0, 128.0)
+
+
+def plan_all_tiers(
+    manifest: TargetManifest, snapshot: ArchitectureSnapshot
+) -> dict[str, TopologyPlan]:
+    """The alpha tier table, recomputed from the pinned artifact and architecture."""
+    return {
+        f"{tier:g}": plan_topology(manifest, snapshot, physical_usable_gib=tier)
+        for tier in ALPHA_TIERS_GIB
+    }

packages/node/meshnet_node/native_backend.py

@@ -0,0 +1,185 @@
+"""Authoritative identity boundary for a native GGUF Shard backend.
+
+The native loader owns the facts about the mapped artifact.  This module turns
+that immutable report and separately pinned deployment inputs into the one
+``ShardIdentity`` DGR-003 permits a native worker to emit.  It is intentionally
+not an adapter for the legacy Transformers backend: that backend has no
+authoritative immutable GGUF artifact pin and must remain identity-free.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+from .native_protocol import BUNDLE_VERSION, SCHEMA_VERSION, pb
+from .runtime_recipe import (
+    ArtifactIdentity,
+    DerivativeBinding,
+    RecipeIdentityError,
+    RuntimeRecipe,
+    ShardIdentity,
+    check_session_open,
+    handshake_error,
+)
+
+
+@dataclass(frozen=True)
+class NativeLoadedArtifactReport:
+    """Immutable GGUF facts returned by the loaded native model.
+
+    The report is copied from ``llama_model_meshnet_range_report`` plus the
+    parsed GGUF metadata while the model is live.  Byte counts are operational
+    evidence rather than compatibility axes, but keeping them beside the range
+    prevents a caller from substituting an unverified range declaration.
+    """
+
+    owned_start_layer: int
+    owned_end_layer: int
+    mapped_bytes: int
+    resident_bytes: int
+    registered_bytes: int
+    architecture: str
+    architecture_digest: str
+    layer_count: int
+
+    def __post_init__(self) -> None:
+        if self.owned_start_layer < 0 or self.owned_end_layer <= self.owned_start_layer:
+            raise RecipeIdentityError("native report has an invalid owned layer range")
+        if self.layer_count < 1 or self.owned_end_layer > self.layer_count:
+            raise RecipeIdentityError("native report range is outside GGUF layer metadata")
+        if min(self.mapped_bytes, self.resident_bytes, self.registered_bytes) < 0:
+            raise RecipeIdentityError("native report byte counts must be non-negative")
+
+
+@dataclass(frozen=True)
+class ImmutableArtifactPin:
+    """Deployment-supplied immutable bytes pin, never inferred from a model name."""
+
+    artifact_id: str
+    revision: str
+    content_digest: str
+    derived_from: DerivativeBinding | None = None
+
+
+@dataclass(frozen=True)
+class NativeNumericalRecipe:
+    """Immutable numerical inputs selected for this native worker instance."""
+
+    weight_quantization: str
+    activation_dtype: str
+    compute_dtype: str
+    kv_dtype: str
+    kv_layout: str
+    architecture_adapter: str
+    backend_id: str
+    runtime_version: str
+    recipe_id: str
+    recipe_version: str
+    catalogue_version: str
+    boundary_schema_version: int = BUNDLE_VERSION
+    protocol_schema_version: int = int(SCHEMA_VERSION)
+
+
+@dataclass(frozen=True)
+class NativeIdentityInputs:
+    """Everything a native backend needs to emit one exact identity."""
+
+    loaded_artifact: NativeLoadedArtifactReport
+    artifact_pin: ImmutableArtifactPin
+    tokenizer_revision: str
+    numerical_recipe: NativeNumericalRecipe
+
+
+def shard_identity_from_native_report(inputs: NativeIdentityInputs) -> ShardIdentity:
+    """Derive identity only from the native report and immutable pinned inputs."""
+    report = inputs.loaded_artifact
+    pin = inputs.artifact_pin
+    recipe = inputs.numerical_recipe
+    artifact = ArtifactIdentity(
+        artifact_id=pin.artifact_id,
+        revision=pin.revision,
+        content_digest=pin.content_digest,
+        architecture=report.architecture,
+        architecture_digest=report.architecture_digest,
+        layer_count=report.layer_count,
+        derived_from=pin.derived_from,
+    )
+    return ShardIdentity(
+        artifact=artifact,
+        recipe=RuntimeRecipe(
+            weight_quantization=recipe.weight_quantization,
+            activation_dtype=recipe.activation_dtype,
+            compute_dtype=recipe.compute_dtype,
+            kv_dtype=recipe.kv_dtype,
+            kv_layout=recipe.kv_layout,
+            tokenizer_revision=inputs.tokenizer_revision,
+            architecture_adapter=recipe.architecture_adapter,
+            backend_id=recipe.backend_id,
+            runtime_version=recipe.runtime_version,
+            boundary_schema_version=recipe.boundary_schema_version,
+            protocol_schema_version=recipe.protocol_schema_version,
+            recipe_id=recipe.recipe_id,
+            recipe_version=recipe.recipe_version,
+            catalogue_version=recipe.catalogue_version,
+        ),
+        shard_start=report.owned_start_layer,
+        shard_end=report.owned_end_layer,
+    )
+
+
+class NativeSessionRejected(RecipeIdentityError):
+    """A native worker refused a ``SessionOpen`` before allocating session state."""
+
+    def __init__(self, error: "pb.ShardError") -> None:
+        super().__init__(f"native SessionOpen rejected: {error.code}")
+        self.error = error
+
+
+class NativeWorkerBackendAdapter:
+    """Small backend-facing adapter around the native loaded-artifact seam."""
+
+    def __init__(self, identity_inputs: NativeIdentityInputs) -> None:
+        self.identity_inputs = identity_inputs
+        self.identity = shard_identity_from_native_report(identity_inputs)
+
+    @property
+    def loaded_artifact_report(self) -> NativeLoadedArtifactReport:
+        return self.identity_inputs.loaded_artifact
+
+    def check_session_open(
+        self,
+        opened: "pb.SessionOpen",
+        *,
+        expected_route_session_id: str | None = None,
+        expected_route_epoch: int | None = None,
+    ) -> None:
+        """Reject incompatible/stale opens at the native worker boundary."""
+        mismatches = check_session_open(
+            self.identity,
+            opened,
+            expected_route_session_id=expected_route_session_id,
+            expected_route_epoch=expected_route_epoch,
+        )
+        error = handshake_error(mismatches)
+        if error is not None:
+            raise NativeSessionRejected(error)
+
+    def on_session_open(
+        self,
+        opened: "pb.SessionOpen",
+        *,
+        expected_route_session_id: str | None = None,
+        expected_route_epoch: int | None = None,
+    ) -> "pb.SessionAccepted":
+        """The native worker's SessionOpen boundary, before session allocation."""
+        self.check_session_open(
+            opened,
+            expected_route_session_id=expected_route_session_id,
+            expected_route_epoch=expected_route_epoch,
+        )
+        return pb.SessionAccepted(
+            schema_version=SCHEMA_VERSION,
+            route_session_id=opened.route_session_id,
+            route_epoch=opened.route_epoch,
+            fingerprint=self.identity.fingerprint.to_proto(),
+        )