5177db25b0
Real ShardRuntimeServicer process bound to a real localhost socket, driven by a generated ShardRuntimeStub over grpc.insecure_channel from a separately spawned subprocess. Proves direct-hop and opaque-relay (exact captured request bytes re-sent, no reinterpretation) produce byte-identical server responses, cross-checked against an independent server-side wire capture. Fails closed on the required negative paths: stale route epoch, expired deadline, malformed/non-tiling fragments, checksum failure, exhausted flow-control credit (with in-band top-up), duplicate idempotency steps (acked, not re-applied), and cancel — both in-band CancelSignal (single work item vs whole session) and the out-of-band unary Cancel RPC, including a Cancel that races ahead of SessionOpen. Supersedes the earlier in-memory fake-seam approach for this ticket, which a policy audit rejected under the no-fake-data rule; that code is not reintroduced. Evidence README rewritten to describe the actual files. 11 passed in tests/test_shard_runtime_harness.py.
Distributed GGUF Runtime evidence
Specification status: planning artifacts only. No distributed GGUF runtime is implemented. DGR-017 cleanup is complete; no runtime implementation story has completion credit.
prd.jsonis authoritative.
Authority and classes
Evidence supports but never overrides prd.json. Valid classes are model-free, fixture, real-model, real-hardware, and release; lower classes cannot satisfy higher-class acceptance. Legacy DGR-001..016 directories remain unchanged for DGR-017 provenance audit and confer no completion credit.
Future story layout
Each DGR-017..071 story writes evidence/<ID>/README.md with summary, exact changed files, exact commands and real outputs, limitations, compatibility/migration notes, and dependent-story handoff. Machine-readable contracts, manifests, metrics, and raw logs live beside it. Never fabricate output.
Real runs record exact model SHA and all split hashes, tokenizer, quant/recipe, llama.cpp pin+patch identity, backend/driver/toolchain, host/hardware/network, commands/environment (without secrets), raw parity/performance/resource results, and evidence class. Models live on configured mounted-drive storage, never /home.
Routing certification records prove only the exact exercised backend/model/recipe lane. Compile-only, fixture, failed, or unavailable lanes remain registered-dark. V4 cache/state evidence must show KV and CSA/HCA/SWA/indexer/compressor data remain shard-local/session-keyed; route recovery evidence must show cache miss plus re-prefill/restart, not migration.