docs(rag): record refuted lexical probe collapse (#098)

[BigSimmo]

Motivation

• The search retrieval budget harness (#098) showed match_document_chunks_text_v2 and match_document_table_facts_text_v2 were each being called ×3 per search, which raised a latency/round-trip question and prompted an attempted optimisation.
• The intent was to collapse near-duplicate lexical sibling variants into a single canonical lexical RPC to reduce Supabase round trips while preserving recall, but retrieval behaviour is a protected RAG surface that requires live canary validation.

Description

• An experimental change collapsed lexical sibling-variant fan-out so only the canonical expanded variant was issued (temporary edits to src/lib/rag/rag-candidate-sources.ts and related tests), but that change was reverted after the live canary showed regressions.
• The revert was applied immediately and the branch pushed, so retrieval behaviour remains unchanged in runtime code; this PR records the evidence and outcome in docs only.
• Updated docs/outstanding-issues.md to document the failed experiment, the canary evidence, the immediate revert, and the follow-up next steps for the route-level POST /api/search budget.
• Files touched: docs/outstanding-issues.md (final committed change); experimental edits and test adjustments were tested and then reverted in the branch history to preserve the audit trail.

Testing

• Ran focused unit tests for the budget/variant harness with npm run test -- --run tests/rag-variant-early-exit.test.ts tests/search-round-trip-budget.test.ts and those focused tests passed (2 files passed, 4 tests passed).
• Ran offline RAG fixture validation with node scripts/eval-rag-offline.mjs, which passed its provider-free fixture checks, and then executed three live eval canaries: baseline canary (passed), experimental canary 30579804611 (failed: document recall @5 fell 1.0 → 0.9444, content recall @5 fell 1.0 → 0.9699, 4 failing cases), and confirmation canary 30580564419 after revert (passed).
• Ran npm run format and repository static/cheap gates; static checks, lint and typecheck passed but a full unit-suite run encountered pre-existing mock-related failures (not caused by this revert) flagged by missing openAISafetyIdentifier mocks, which remain outside the scope of this PR.
• No production changes were deployed and provider-backed eval dispatches were run only with explicit approval as recorded in the audit trail; the PR contains documentation of the experiment, evidence, revert, and recommended next steps.

---

Codex Task

[coderabbitai]

Warning

Review limit reached

You’ve reached a temporary PR review limit under our Fair Usage Limits Policy.

Your recent review volume is higher than typical usage, so adaptive limits are currently applied.

Next review available in: 59 minutes

Your organization has reached its usage spending cap. Adjust your spending cap in the billing tab.

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: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: ea6cdd4e-137b-4220-9c73-8f6f4e70e409

📥 Commits

Reviewing files that changed from the base of the PR and between 2e670ee and 11b19c0.

📒 Files selected for processing (2)

• docs/branch-review-ledger.md
• docs/outstanding-issues.md

---

_{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: b14adae276

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".

[BigSimmo]

Temporarily paused during the owner-authorized ordered merge sweep so main cannot move between exact-head review and merge. The branch and all commits are preserved; this PR will be reopened at its verified place in the queue.

[BigSimmo]

@cursoragent this PR will be re-opened soon. Please resolve any failing CI, run bugbot and resolve any bugbot comments against current repo and fix any merge conflicts and also fix any errors and address and fix all comments or suggestions that are likely to improve the PR and Codebase. Please implement smallest required fix according to best code practices. Check your work. Leave this PR closed for now but still make all these changes and run the review. I want it ready to go for when I re-open it.

[cursor]

PR #1503 is reopen-ready and still CLOSED.

Fixed

• Codex P2: #098 next-action (b) no longer invites collapsing the ×3 lexical probes. It now bans wholesale collapse (REFUTED by canary 30579804611) and points to different latency work that keeps all three variants.
• Merge conflict: rebuilt docs/outstanding-issues.md from current main, re-applied only the #098 update (also dropped the duplicate Done #1450 block).
• Main sync: branch tip ed4afc44 is 0 behind / merge-tree clean.
• Review thread: Codex comment replied + resolved.

Review / Bugbot

• No Bugbot findings; docs-only delta.
• No remaining high-confidence defects.
• Ledger recorded under pr-1503 reopen-ready.

Checks

• format:check — All matched files use Prettier code style!
• check:outstanding-issues — guard passed
• check:branch-review-ledger — guard passed
• Hosted CI won’t refresh on a closed PR; prior closed-head CI was green for this docs-only scope. CI will re-run on reopen.

Note

GitHub may still show the old closed headRefOid until you reopen; the branch tip is ed4afc44 with the fix.

  

[supabase]

This pull request has been ignored for the connected project sjrfecxgysukkwxsowpy because there are no changes detected in supabase directory. You can change this behaviour in Project Integrations Settings ↗︎.

---

Preview Branches by Supabase.
Learn more about Supabase Branching ↗︎.

[BigSimmo]

@cursoragent This PR was just reopened. Do not merge, force-push, rebase, or arm auto-merge
unless I explicitly ask.

1. Confirm GitHub is on the current branch tip (not the old closed head).
   Snapshot head SHA, behind/ahead, mergeable state, required checks, and
   unresolved actionable threads.

2. Mergeability on this tip only:

   • Real conflict (merge-tree dirty): smallest fix, push.
   • Behind-but-clean: if required CI is in flight, wait and sync once late;
     otherwise sync once now. Prefer human/operator identity for update-branch;
     else git merge origin/main + push. No sync thrash.
   • Clean/current: leave it.
     Missing checks while mergeability is dirty ≠ green.

3. Fresh required CI on this tip only — ignore pre-close rollup. On failure:
   smallest fix, push, re-check. Don’t rerun unless a fix landed or a check
   is stuck. No provider-backed gates without approval.

4. New/unresolved actionable Codex/Bugbot/Copilot/human threads: smallest fix
   or short already-fixed/not-actionable reply, then resolve. Ignore
   rate-limit/summary noise.

5. Stop when: merge-tree clean, no actionable threads, required checks green
   or clearly in progress on this tip. Report tip SHA, sync done/not needed,
   CI status, threads handled, residual risks. Leave the merge decision to me.

[cursor]

1. Snapshot (at reopen)

• GitHub head matched branch tip 5135b439 (not the old closed head).
• Behind 12 / ahead 6; mergeable=CONFLICTING, DIRTY.
• merge-tree dirty on docs/outstanding-issues.md.
• Required CI missing because merge ref could not build (stale conflict signal).
• Unresolved actionable threads: 0 (Codex P2 already resolved).

2. Sync

Real conflict → merged origin/main once and pushed.

Conflict fix (smallest correct):

• Rebuilt docs/outstanding-issues.md from main (kept #093 archived + main’s #098 canary-attribution Note).
• Re-applied only this PR’s intentional next-action (b) ban on wholesale sibling collapse (no should be collapsed invite; no disputed Refuted canary paragraph).

3. CI on tip 11b19c04

All required checks SUCCESS on this tip:
PR mergeability, PR policy, Change scope, Static PR checks, PR required (+ Semgrep/Gitleaks).

4. Threads

None new/unresolved. Rate-limit/summary noise ignored.

5. Stop state

Item	Status
Tip	11b19c0473df833f2ce135e7f152b6070168c1fa
Sync	done (conflict resolved)
merge-tree	clean; 0 behind
GitHub	MERGEABLE / CLEAN
Threads	0 actionable
CI	required green on tip

Residual risk: PR body still narrates canaries 30579804611 / 30580564419 as collapse evidence; main’s corrected #098 Note says those runs were on unrelated SHAs. Row text now matches main’s correction + the collapse ban; body prose was left unchanged.

Merge decision left to you — no merge / force-push / rebase / auto-merge.