docs(issues): correct #018/#019 root causes from run #61 artifact; add #029/#030

[BigSimmo]

What

Corrects the recorded root causes for #18 and #19, and files two new findings (#29, #30). Docs-only — docs/outstanding-issues.md.

Why

Both items were filed as retrieval problems. Mining the run #61 --dump-answers and golden-retrieval artifacts (zero spend, read-only) shows both are answer-stage defects, with healthy retrieval in each case.

#18 — not retrieval-depth

Retrieval is grounded=true, 4–6 citations, high_confidence_extractive_retrieval. The answers are:

• quality-lithium-monitoring-range / quality-adhd-medication-monitoring → the source_backed_review_fallback boilerplate stub, with answer_sections: []
• quality-antipsychotic-metabolic-monitoring → a mangled 73-char fragment: "The guidance for metabolic is that compliance, monitoring and evaluation."

The prior note — "their live chunk sets apparently surface no admissible schedule sentence" — was an inference, not a measurement, and is falsified: the chunk sets were never exercised because no answer was composed.

Also recorded: do not widen the eval matcher. The stub contains monitoring/maintenance/level/range, so widening rag-eval-cases.ts:192 to production vocabulary would make these stubs falsely pass. The narrow matcher is doing its job.

#19 — not comparison-class retrieval coverage

Golden retrieval for this exact case: documentRecallAt5=1, reciprocalRankAt10=1, missingDocumentSubstrings=[], failures=[], with "Admission of Community Patients (AKG).pdf" at raw rank 4–5, satisfying the strict AdmissionCommunityPts alias.

The answer stage then drops it from answer.sources top-5 and promotes "Falls Prevention and Management (AKG).pdf" — raw rank 8, off-topic — in its place. Suspect: balanceCrossDocumentResults / rankAnswerEvidence.

Separately noted: nothing guarantees both sides of a comparison are represented — comparison_multi_document_gate (rag.ts:2043) is satisfied by any 2 distinct docs, including two discharge-side ones.

New items

• chore(deps-dev): bump esbuild from 0.28.0 to 0.28.1 in the npm_and_yarn group across 1 directory #29 (P2) — 12 of 30 answer-quality cases return the fallback stub; targeting masks it for keyword-only intents (contraindication/document_lookup), so the metric under-reports the problem.
• Potential fix for code scanning alert no. 1: Workflow does not contain permissions #30 (P3) — wide-tier alias lets one document satisfy both comparison slots (latent false-pass).

Scope / safety

• No code fix included. Both root causes sit on protected RAG surfaces and require a provider-backed eval/canary pair plus explicit approval. Shipping an unvalidated fix here is precisely the "offline-green is insufficient" trap.
• Docs-only → no RAG impact: line required (pr-policy keys on code paths).
• Gates: prettier --check ✓, docs:check-links ✓ (1104 refs resolve).

🤖 Generated with Claude Code

[coderabbitai]

Important

Review skipped

Review was skipped as selected files did not have any reviewable changes.

💤 Files selected but had no reviewable changes (1)

• docs/outstanding-issues.md

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: f34c87ed-bebf-44bb-971c-ac7fd3e70d92

📥 Commits

Reviewing files that changed from the base of the PR and between 46f143d and 12d7a1a.

📒 Files selected for processing (1)

• docs/outstanding-issues.md

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch claude/rag-018-019-root-cause

---

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

[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 ↗︎.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 70b10635cd

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".