Two anchor-vs-reality drifts found during /read-mem audit: 1. start_go_stack.sh never started validatord :3221, even though it shipped 2026-05-02 (f9e7241) and STATE_OF_PLAY claims it as part of the persistent stack. Cold-boot quietly omitted it, leaving /v1/iterate unreachable on the persistent gateway. Fix: factored chatd's conditional-start block into a start_shared helper, called for both chatd :3220 and validatord :3221. Same shared-with-smokes posture as chatd (no S3 / JSONL-only state, no temp-toml override needed). 2. STATE_OF_PLAY header claimed 3 parity probes / 32 assertions. Reality is 6 probes / 38 assertions since subject_audit landed in 262a77a (2026-05-03). Header refreshed; cross-references the three runtime-divergence classes documented at lakehouse/STATE_OF_PLAY.md lines 36-39. Parity reports regenerated as verification artifact (all 6 still green: 8+12+2+4+1+6). Same pattern as c0a55b1. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
727 B
727 B
/v1/embed cross-runtime parity probe
Date: 2026-05-05T08:22:55Z
Rust: http://127.0.0.1:3100/ai/embed · Go: http://127.0.0.1:4110/v1/embed
Model: nomic-embed-text (forced — overrides each side's default)
Match metric: cosine similarity ≥ 0.99999
Identical text → both endpoints. Cosine compares vector DIRECTION (the operationally-meaningful property for HNSW retrieval); byte-equal isn't expected because Go round-trips through float32 internally while Rust stays at f64.
Tally: 8 match · 0 diff (out of 8 fixtures)
Cosine ≥ 0.99999 on every fixture — embed parity holds post-sidecar-drop. Rust and Go produce vectors that point in the same direction in 768-dim space.