[Dong1017]

Scope and decision

Evidence-only follow-up to #255. No production runtime or corpus changes; the single experimental weighted-token-v2 candidate is retained only as a patch/Git bundle for reproducibility. Do not promote the candidate or close #255.

The prospectively frozen development gate failed:

Partial-label metric	Baseline	Candidate
Hit@5	4/6	5/6
MRR@5	0.583333	0.541667
Complete relevant deliveries	4	5

The Hit gain does not override the MRR regression. Evaluation stopped without tuning. The six heldout cases were not exposed to the candidate implementer and were not evaluated; case-author/source-review access for preparing labels is acknowledged. Their contents are not included here.

Artifacts

• Readable per-case ranks, output/context bytes, decision and limits: report.
• Accepted protocol, development inputs, complete lossless gzip outputs, raw hashes, default-output parity captures and candidate patch/bundle.
• Evidence inventory adds 33 hashes (232 total, 11 bundles). Original campaigns remain unchanged.
• Historical experiment baseline: 1147c5bb41724826b64ad13cebc8a4dce5cc278b; rejected candidate: 55a458328195335586eb6c6c1301ba34a8218e63.

Verification and limits

• Candidate focused compatibility/synthetic tests: 41 passed; six baseline/candidate default serialized outputs identical.
• Evidence-only tree affected inventory tests: 17 passed; ruff check src tests tools passed.
• Independent artifact audit verified integrity, gzip outputs, patch/bundle, and all 38 emitted contexts per arm against complete required resources, exact bytes/hash/revision.
• Current hosted CI is pending. Local tests/artifact review do not imply model efficacy, speed nonregression, or Issue acceptance.
• No new timing campaign, model/service calls, hardware experiments or spending. No merge, formal approval or Issue closure requested by this PR.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-30T01:45:55.344706Z	bdfee4e	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[Dong1017]

Scoped LGTM for this evidence-only delivery at bdfee4e against main 1147c5b. No production source changes or candidate promotion.

Independent verification reproduced the retained development arithmetic: Hit@5 4/6 to 5/6, MRR@5 7/12 to 13/24; the frozen nondecreasing-MRR gate therefore FAILS. All 38 emitted contexts per arm were checked for complete required resources and exact bytes/hash/revision; saved default outputs match. Publication audit verified 33 hashes, 16 compressed outputs and the exact retained candidate patch/bundle. Provenance wording now distinguishes source-review access from heldout exposure: heldout was not exposed to the candidate implementer and was not evaluated. No tuning or new timing campaign occurred.

Affected evidence tests passed (17); current Ready CI https://github.com/InferMatrixLab/InferRSI/actions/runs/40987495415 passes both full-check and installed-package Python versions. The exact-head native review completed with no findings and no inline threads remain. The earlier draft-trigger run was superseded by this successful Ready run.

This accepts honest, reproducible evidence delivery only. Accuracy improvement was not established, runtime remains unchanged, heldout remains unevaluated, and #255 stays open. No model/hardware efficacy claim, formal APPROVE or tracker closure.