docs: freeze measured embedding benchmark result

[stone16]

Closes #128

Outcome

• replaces the tracked pending_corpus placeholder with the reviewed,
  aggregate-only 30-case benchmark result and exact run identity;
• records Qwen as the frozen Pareto winner: 17/30 (0.5666666667) versus
  multilingual-e5-small at 13/30 (0.4333333333) across case hit, macro
  Evidence recall, and micro Evidence recall;
• records the explicit win over the standing 3.8% twin baseline and links the
  activated-profile acceptance: Qwen 7/28 (0.25) versus twin 0/28 (0.0);
• updates the evaluation status and freezes the exact aggregate artifact under
  the existing privacy test; the full report and per-slice material remain in
  the ignored .context-engine root;
• closes two post-merge fix: bounded micro-batch local embedding and one-hop multi-anchor exclusion #230 review residuals: exact one-fragment micro-batch
  naming and discoverable disclosure of the temporary timeout refusal-category
  conflation.

Definition of Done mapping

1. Resolved identity refusal — test_embedding_benchmark_identity.py.
2. Closed report schema and required identity/metrics —
   test_embedding_benchmark_report_schema.py.
3. Canonical [M1] Golden set v1 + layered judges #129 retrieval judge and hand-checked metrics —
   test_embedding_benchmark_metrics.py and
   test_embedding_benchmark_retrieval_judge.py.
4. Deterministic provider runs — test_embedding_benchmark_determinism.py.
5. Runtime composition/dimension untouched —
   test_embedding_benchmark_runtime_untouched.py; this diff changes neither.
6. Aggregate-only frozen record —
   test_embedding_benchmark_report_privacy.py freezes the exact object and
   rejects query/title/path/excerpt keys and personal paths.
7. Provider vector validation — test_embedding_provider_validation.py.
8. Tracked CLI, help, and synthetic execution —
   test_embedding_benchmark_cli.py; fresh --help also passed.
9. Real-model run — tracked record now carries the exact dataset digest, run
   identity, top-k 10, 4,166-document count, 30-case metrics, and timings from
   the schema-valid ignored report.
10. Actual verdict — tracked record names Qwen the winner under the frozen
    Pareto rule, states the win against the 3.8% twin baseline, and records the
    activated online 7/28 versus 0/28 reference.
11. Fresh verification — all required commands below pass, including the
    unchanged M0 veto on the preserved dogfood database at schema
    20260803_0055.

All issue DoD boxes are evidenced; there is no remaining closure blocker.

Verification

• focused benchmark validation: 118 passed
• Qwen provider unit tests: 9 passed
• bounded-document-refusal integration test: 1 passed
• uv run context-engine-embedding-benchmark --help: PASS
• make lint: PASS
• make typecheck: PASS (mypy 513 sources plus TypeScript)
• make test: PASS (2,619 tests)
• make security-gate: M0 SECURITY PASS (187 retained tests; repository and
  live database both at 20260803_0055)
• git diff --check: PASS

No production Runtime composition or embedding dimension constant changes.

[coderabbitai]

Warning

Review limit reached

@stone16, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 7 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 86466e5e-a8ff-40e8-a01d-b7040c34d2b4

📥 Commits

Reviewing files that changed from the base of the PR and between c709740 and a406fb3.

📒 Files selected for processing (5)

• adapters/embeddings.py
• docs/evaluation/2026-07-29-embedding-benchmark.json
• eval/README.md
• tests/unit/test_embedding_benchmark_report_privacy.py
• tests/unit/test_local_qwen_embedding_provider.py

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a406fb3730

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Replace the hard-coded acceptance URL

stometa, when the repository moves or the linked issue comment becomes unavailable, this frozen verdict loses its only pointer to the acceptance evidence. Store a stable repository-owned evidence reference instead of embedding the volatile GitHub URL, as required by the repository charter.

AGENTS.md reference: AGENTS.md:L94-L96

Useful? React with 👍 / 👎.