[Dong1017]

Summary

Related tracker: #255; its broader requirements remain open.

• Add explicit relevance_method="weighted-raw-string-v1" / --relevance-method weighted-raw-string-v1 so actual string newlines/tabs do not become phantom lexical matches from JSON escapes.
• Preserve the legacy default, substring/casefold matching, weights, nonstring JSON behavior, explicit IDs, filters, budgets and optional-context scoring.
• Publish the unchanged frozen evidence, an independent audit and a runnable cue → skill/context demo.

Evidence, checks and limits

The one frozen synthetic contract study passes all 14 scalar cases and 2 ordering fixtures: five phantom field-token contributions removed; 15 original-string and 11 expected nonstring-JSON positives retained; 21 compatibility requests preserved. Actual newline+tab title: score 8→0; literal backslash-letter title: 8→8. This is not general retrieval accuracy, efficacy or latency evidence.

Frozen baseline 1052925f, candidate 728984c5; archive and original Git bundle preserve those identities. Publication integrates onto 436c950c without rerunning the study. Independent source, retained-output and publication-wrapper reviews found no blocker.

• Frozen-candidate full CPU suite: 2707 passed, 12 skipped; 284 subtests passed.
• Current-main affected suite: 108 passed, 2 skipped.
• Publication inventory: 17 passed, 321 hashes verified; Ruff clean.
• All four hosted Ready jobs passed on this exact head: Python 3.10/3.11 full checks and clean installed-package checks. Current-head knowledge-model-policy passes.

Evidence and provenance distinguish this new configured author from historical protocol/fixture/reviewer records. No historical PR authorship is certified.

{
  "schema_version": 1,
  "head_sha": "95a35192668d155e30c5b1d97146a6a5ddc84d96",
  "human_only": false,
  "material_ai_authors": [
    {
      "tool": "ChatGPT Work collaboration agent /root/implement_raw_string_contract",
      "model_id": "gpt-6-astra",
      "effort": "xhigh",
      "role": "author"
    }
  ]
}

[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 2026-10-30T10:48:42.599950Z	95a3519	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]

LGTM for the opt-in raw-string substring contract at 95a35192. The mode removes serialization-created phantom matches while preserving substring matching and the unchanged default path.

• Frozen evidence: 14 scalar cases and 2 ranking fixtures pass; 5 phantom contributions removed, all 26 expected positives retained. These are synthetic contract checks, not general retrieval-accuracy evidence.
• 21 compatibility requests preserve behavior. Current Ready CI and author declaration pass; native review has no findings, and no unresolved review threads remain.
• The new main guidance change has no file or reader/ranker/corpus intersection. The broader Indexer tracker remains open, with no default promotion.

Evidence and integration limits

Head: 95a3519. Frozen candidate: 728984c524cf696d063ff7ef0688ae126c8546a6. Local verification: 2707 passed, 12 skipped; current-base patch integration: 108 passed, 2 skipped; Ruff passed. Ready CI, declaration, native source comment 6520834588. Earlier opened-event run was superseded by successful Ready CI. CI tested base 436c950; subsequent main db4e8a1 was inspected for integration intersections, not a fresh scored-study run. Frozen evidence was retained unchanged.