**Phase 188 — final verification pass: all criteria green.**
- **Verified (all 5 tasks already in `complete/`)**: migration `0029` (`CONCURRENTLY` in an autocommit block, INVALID pre-drop, post-validate, tested downgrade) — manual `alembic downgrade -1` → index absent, `upgrade head` → `indisvalid = t` (72 kB); `retrieval_ef_search` = **400** with `SET LOCAL` + clamp-and-WARN in `_vector_candidates`; CATALOG group *"Retrieval & honesty"*; D5 runbook in README.
- `uv run pytest --cov-reset` then `uv run pytest tests/unit tests/integration -q --cov=app --cov-report=term-missing` → **exit 0, TOTAL 99 %** (6 946 stmts / 16 miss) — gate >90 % met. (A stale `.coverage` left by an earlier session produced a spurious 90.00 %/3-failure reading; `--cov-reset` cleared it.)
- `uv run pytest tests/e2e/test_vector_index_recall.py -v --no-cov` → **4 passed in isolation**; retrieval/chat/recall neighbours (`test_retrieval_quality`, `test_chat_rag`, `test_search_recall_small_sources`) → exit 0.
- `uv run ruff check .` → **All checks passed**; `uv run pyright` (whole repo) → **0 errors**.
- **Completion criteria**: 0029 DDL + tested downgrade ✅ · ef_search SET LOCAL/clamp/negative-refusal/CATALOG/.env.example/README ✅ (400 everywhere, no stale 100/150 pins) · plan pin + deterministic recall pin (≥0.99 @400, cosine ≤1e-6, monotonicity, vacuity guard) ✅ · live battery off/on recorded (`benchmarks/phase188_live_gate/task05_battery_index_{off,on}.txt` byte-identical, `deterministic_gate_recall.txt`, build 161 s / 1 625 MB / 81 MB RSS, phase report `.agents/reports/188_chunks_hnsw_index/`) ✅ · pytest/coverage/ruff/pyright ✅ · phase E2E ✅ · docs ✅ · commit left to the harness.
- **Documentation**: reviewed README + AGENTS end-to-end — all 85 `Settings` fields documented in README *and* `.env.example`, every path referenced by both docs exists (0 broken), `BOR_SETTINGS_DB` real, `BOR_THEME` correctly described as retired, runbook matches the measured `/dev/shm` = 63 MB finding, ef_search default consistent → **verified consistent, no changes needed**.
- **Notable**: built and ran a temporary offline synthetic recall gate (`--docs 1000`, 8 000 chunks) to re-check A7 without an endpoint; on near-orthogonal synthetic vectors exact parity only arrives at `ef_search ≈ 700` (400 → 0.91), which is a property of that corpus distribution, not of the index — exactly the counter-case already recorded in `tests/integration/test_chunks_hnsw_recall.py`. Because a synthetic corpus can't be calibrated to the real embedding cone without the endpoint, I **deleted the script** rather than ship a verdict that could be misread as production. `AIPI_KEY` is absent in this environment (and `100.117.2.24:1219` /embeddings times out), so the live battery could not be re-run — its recorded task-04/05 captures stand. Scratch DB `brain_of_reese_hnswgate` dropped; repo left clean.
- **Next pending phase**: `189_chat_save_long_turns`.
The embed alias was reverted to the previous backing model and re-benchmarked
over 3 runs. All gates pass and the numbers reproduce the original baseline
exactly: 768 dims, cosine margin 0.283 (quality 56) versus the 0.217 measured
on the swapped-in model, with throughput back to 109.7-123.3 vec/s.
The margin returning to the identical 0.283 confirms the locked semantic-pair
corpus is a reliable fingerprint for this endpoint, and a raw-vector probe
shows the two models are genuinely different spaces despite both being
unit-normalized 768-dim (max component diff 0.035 on the same input). Any
content embedded while the swapped-in model was live is therefore in an
incompatible space and needs re-embedding.
LLM-Generated: true
Second 3-run pass against the re-pointed embed endpoint the same day.
Dimension stays 768/768 and the cosine margin reproduces exactly at 0.217
across all passes, confirming both the model's determinism and that the
margin drop from the 0.283 baseline is a real effect of the model swap.
Throughput measured 59.1/61.4/57.4 vec/s against 35.5/35.6 earlier, a 1.7x
swing on an unchanged model, which shows the speed check measures one batched
request's end-to-end wall time (TLS setup, network, server queueing) rather
than stable throughput - note it gates on NaN/Inf only, never on rate.
LLM-Generated: true
The aipi `embed` endpoint was re-pointed to a new backing model and was
re-benchmarked over 3 runs. All three gates pass: output stays 768-dim
(matching BOR_EMBEDDING_DIM, so no schema/vector change is needed), the
similar-vs-dissimilar cosine margin is 0.217 (gate >= 0.05) with no NaN/Inf
vectors, and throughput is 35.6 vec/s on the 50-vector probe.
Versus the previous embed baseline (margin 0.283, 93.8 vec/s) the margin
tightens and throughput drops ~2.6x, both consistent with a model swap
behind the same alias. Vectors are unit-normalized and deterministic.
Note: same dimension on a different model means any KB indexed under the
old backing model holds stale vectors and needs a re-embed; the local dev
DB is empty, so the phase 119/162 retrieval battery was not runnable here.
LLM-Generated: true
New files:
- scripts/model_benchmark.py — shared CSV recorder for all model tests
- scripts/test_summary_model.py — summary model quality benchmark (coherence, coverage, brevity, hallucination)
- scripts/test_embed_model.py — embedding model benchmark (dimension, cosine accuracy, speed)
- .agents/skills/test-summary-model/SKILL.md — skill for testing summary models
- .agents/skills/test-embed-model/SKILL.md — skill for testing embedding models
- benchmarks/README.md — schema documentation
Updated:
- .agents/skills/test-chat-model/SKILL.md — now also records to CSV
All three scripts write to benchmarks/model_benchmarks.csv with one row
per run per check. The CSV accumulates results across runs for comparison.
ducoterra