Improve hybrid ranking with production-source intent

[justrach]

Summary:

• hill-climb ANN RRF and intent-aware path priors on 12 React production-source queries
• demote excluded tests, fixtures, debug/server variants, and docs while preserving explicitly requested material
• apply the same intent to symbol definitions and retain the exact lexical fallback policy
• keep semantic-index concurrency at four after a 6,713-chunk 1/2/4/8 benchmark
• add a reproducible held-out regression harness and measured results

Measured React retrieval:

• train hit@5: 0.625 -> 1.000; MRR: 0.238 -> 0.729; NDCG@5: 0.186 -> 0.646; recall@5: 0.375 -> 0.719
• held-out hit@5: 0.750 -> 1.000; MRR: 0.150 -> 1.000; NDCG@5: 0.226 -> 0.738; recall@5: 0.625 -> 0.750
• test-like top-five pollution: 3.00 -> 0.00 train and 1.25 -> 0.00 held out

Verification:

• zig build test --summary all: all 31 steps succeeded; four intentional skips
• scripts/e2e_mcp_test.py: 76/76 passed
• focused React useState probe ranks ReactHooks.js and ReactFiberHooks.js first, with the production public hook selected as the first definition
• git diff --check passed

[github-actions]

👋 Thanks for the contribution! Quick heads-up: this repo lands changes on the current release/* branch, not main.

Please retarget this PR via Edit → base branch to the active release/* branch shown in the repository branch list.

(Automated hint — reply here if you need a hand.)

[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 2026-08-30T10:42:06.423224Z	845cc68	PR opened

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.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 845cc68eb1

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]

Scope exclusions to the referenced path class

When a task requests server code but separately excludes another class—such as the added fizz-streaming query, which asks about “server rendering” and then says “Exclude tests and fixtures”—the global exclusion_clause and mentions_server flags incorrectly set exclude_server. The desired server files are consequently multiplied by 0.05 in hybrid ranking and omitted entirely from symbol definitions, including every primary gold file for that benchmark query. The replay harness does not reproduce this behavior because it only demotes servers for phrases such as server-only or client implementation, so its reported gains cannot detect the regression; associate each exclusion with its actual object instead.

AGENTS.md reference: AGENTS.md:L19-L19

Useful? React with 👍 / 👎.