audit fixes (P1): medication reference correctness — H1, H2, M22

[BigSimmo]

Summary

Audit remediation package P1 — Medication reference correctness, from docs/audit/full-repository-audit-2026-09-02.md (PR #2573). Findings fixed: H1, H2, M22. Each fix landed test-first as its own commit and was adversarially reviewed by a second agent before the gate ran.

• cdaf9b6 medications: read formulation strengths decimal-first, and never badge a concentration (H1)
• 7005a61 medications: stop a documented interaction composing to the all-clear verdict (H2)
• a298740 medications: bind the lexicon sign-off to the mappings it covers (M22)
• sign the source-side exclusions too, so an alert-routing change invalidates the sign-off (Codex review, P1 — proposed in medications: render the lexicon's source-side exclusions on the review sheet #2582)

Findings in this package not fixed here, with reasons:

• none

Decisions needed from the owner:

• Re-attest the lexicon review sheet: a clinician must re-check the terms that moved since 2026-08-22 (acei now Perindopril + Ramipril, statins three drugs, fibrates two rows to one) and record | Mappings hash | f789156464c90d26b99a81c871db216bdedc2cd1de07a07f3fac9cbe88e0ea54 | in the Sign-off block. Until then the sheet's first line says the sign-off is not current.
• Whether check:medication-lexicon-report should exit non-zero (not just warn) once a hash-bound sign-off exists. Making it fail today would turn every PR red until the re-attestation above, which is why it warns.
• Verdict-band copy for a documented low-severity interaction: it now reads "PATIENT INFORMATION / Review for this patient" above "1 interaction". The existing info-tone copy was reused rather than inventing clinical wording. If the owner wants something explicit such as "Documented, low severity", that is a copy decision (and a distinct tone ranked between caution and success would need src/lib/semantic-tone.ts, which this package does not own).
• Formulation badge for combination and concentration-only products (buprenorphine/naloxone, amoxicillin-clavulanate, levodopa-benserazide, all injectables and inhalers stated per mL): the badge is now absent. Confirm that is preferred over a non-numeric form label such as "Ampoules".

RAG impact: no retrieval behaviour change — medication reference rendering only; no RAG file touched

Verification

• npm run verify:pr-local — heavy plan (check:runtime, check:installed-lock-parity, format:changed, check:diff-integrity, sitemap:check, check:repo-awareness-snapshot, docs:check-index, docs:check-inventory, docs:check-scripts, docs:check-links, check:branch-review-ledger, check:outstanding-issues, check:ledger-write-discipline, lint, typecheck, test, build, check:rag:fixtures, check:medication-interactions, check:medication-lexicon-report); closing summary - failed: (none) · - not reached: (none) (EXIT=0). The unit suite completed inside the run; the reviewer's independent run of the same suite reported Test Files 955 passed / Tests 12367 passed.
• npm run check:diff-integrity — [diff-integrity] PASS — 4 changed test file(s), 62 -> 90 test case(s), against base 705c2f64a.
• npm run check:medication-interactions — up to date (531 rows); npm run check:medication-lexicon-report — up to date (37 catalogue terms) followed by the new sign-off WARNING line, exit 0.
• Focused proof by the reviewer: the four touched test files Test Files 4 passed (4) / Tests 105 passed (105); all 19 medication/badge/cross-mode/prescribing test files 19 passed / 330 passed; eslint and prettier clean on every touched file; typecheck pass.

Verification not run: npm run verify:ui — browser proof left to CI (the pinned Chromium is not installed in this container); no narrowed browser run is claimed as the full gate.
Verification not run: npm run verify:release — no release or handoff confidence is claimed.
Verification not run: provider-backed gates (check:production-readiness, every eval:*) — nothing here touches OpenAI, Supabase, Railway or Sentry at run time; all work was offline.

Risk and rollout

• Risk: medium — clinical-reference behaviour changes, each pinned by a new test and reviewed adversarially. H1 changes 80 formulation labels across the committed corpus: about 50 become the correct strength (varenicline 0.5 mg, carvedilol 3.125 mg, bisoprolol 1.25 mg, dexamethasone 0.5 mg and so on) and about 30 disappear because the only number stated is a per-mL concentration or a combination pair — the conservative loss chosen over a wrong number. H2 changes the visible verdict for the 51 low/none/safe/beneficial interaction rows from a green shield headed "No alert found" to the existing info tone headed "Patient information / Review for this patient". composeMedicationVerdict now degrades info to neutral when the analysis is incomplete so the manual-review headline is never lost (the Never show an all-clear over a medication check that could not run #2538 guard is kept and extended). M22 warns rather than exits non-zero, so the gate stays green while the review sheet itself now says the sign-off is not current; the sheet's human Outcome row was not edited.
• Rollback: revert this PR's commits; each commit is one finding and reverts independently.
• Provider or production effects: None.
• RAG impact: no retrieval behaviour change — medication reference rendering only; no RAG file touched

Clinical Governance Preflight

Each item confirmed against this package's diff (findings H1, H2, M22).

• Source-backed claims still require linked source verification before clinical use
  • No change to source verification, provenance or citation requirements.
• No patient-identifiable document workflow was introduced or expanded without explicit governance approval
  • None introduced or expanded. No patient-data handling, privacy controls or document-access behavior changed.
• Supabase target remains Clinical KB Database (sjrfecxgysukkwxsowpy)
  • No Supabase env value, migration target or configured Supabase project/target changes.
• Service-role keys and private document access remain server-only
  • Service-role credentials stay server-side only, and private document access paths are unchanged.
• Demo/synthetic content remains clearly separated from real clinical sources
  • Unchanged. Demo and synthetic content stay separated from real clinical sources.
• Source metadata, review status, and outdated/unknown-source behavior remain conservative
  • Made more conservative. A documented low-severity interaction no longer reads as an all-clear, and an unbound or stale lexicon sign-off is no longer presented as current.
• Deployment classification/TGA SaMD impact was checked when clinical decision-support behavior changed
  • SaMD/TGA assessment reviewed. The changes correct reference rendering defects the 2026-09-02 audit recorded and do not add decision-support behaviour.

Notes

• Files touched: src/lib/medication-badges.ts, src/lib/medication-interactions.ts, scripts/build-medication-lexicon-report.ts, docs/medication-interaction-lexicon-review.md, tests/medication-badges.test.ts, tests/medication-interactions.test.ts, tests/medication-low-severity-verdict.dom.test.tsx (new), tests/medication-lexicon-report-signoff.test.ts (new), docs/design-system/adoption-manifest.json (regenerated by the pre-commit hook).
• Codex review thread (P1), answered in medications: render the lexicon's source-side exclusions on the review sheet #2582: catalogueMappingsHash hashed only the selector's own deny-list and the resolved targets, so a sourceDenySlugs edit changed which source rows raise an alert while the signed payload stayed byte-identical and a recorded sign-off would still read as current. The live statins and fibrates exclusions are exactly that shape. medications: render the lexicon's source-side exclusions on the review sheet #2582 puts sourceDenySlugs into the signed payload, pins it with a new case in tests/medication-lexicon-report-signoff.test.ts, and moves the sheet's hash from 4365ee59… to f7891564…. Merge medications: render the lexicon's source-side exclusions on the review sheet #2582 into this branch before merging this PR, or the Mappings hash quoted above is the pre-fix one.
• This body's Clinical Governance Preflight was rewritten on 2026-09-03 to use the canonical item text verbatim, with the per-item rationale on a sub-bullet. scripts/pr-policy.mjs matches each required item by exact string, so the previous "item — rationale" one-liners satisfied only 2 of 7 and failed the required PR policy check. Re-evaluated locally against the committed evaluator: ok: true, errors: [], warnings: [].
• Reviewer verdict (round 1): pass. Highlights: pre-fix behaviour confirmed from origin/main (the old regex yields "5 mg tablet" for varenicline and "5 mg" for the buprenorphine/naloxone film), so every new case fails without the fix; a corpus-wide invariant now asserts a badged number must be a whole strength token that is not a concentration; info sits below success in SEMANTIC_TONE_PRIORITY, but no current code path can emit a success candidate beside a documented interaction, so the gap is latent — a dedicated tone ranked above success is the owner-decision item above. Pre-existing, outside this finding: list-tail shapes still badge the last listed strength (morphine "Tablets IR (10, 20 mg)" → "20 mg tablet"), which is a real strength and was the same before this branch; worth a follow-up finding.
• One owner per file across the remediation programme: this package's files are edited by no other open remediation PR.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TXW5uh3xxxJWunb88yN6ha

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Team

Run ID: 058e7ce7-33fa-425f-ab8c-cfa14fd3ec1d

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{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 ↗︎.

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_7fbe73b3-758e-4205-b15f-77c901ee1ee7)

[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 2026-09-03T00:12:54.325567Z	a298740	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.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a29874017f

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".

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_1227a24d-034c-4103-a6fc-998b353a8123)

[Copilot]

🟢 Approval recommended

The clinical-safety behavior changes are well-scoped and heavily test-pinned, with only minor wording/test-robustness nits identified.

Pull request overview

Implements the P1 audit remediation package for medication reference correctness (findings H1, H2, M22) by tightening medication strength badge parsing, preventing documented low-severity interactions from composing into an “all-clear” verdict headline, and binding lexicon clinical sign-off to the exact catalogue mappings via a stable hash (with supporting tests).

Changes:

• Fix formulation strength badges to preserve decimal strengths and avoid badging concentrations (mg/mL, mg/2 mL) or combination ratios (e.g. 500/125 mg).
• Re-map documented low/none/safe/beneficial interaction severities away from success and ensure incomplete-check “manual review” messaging still outranks info.
• Add lexicon report sign-off binding via a SHA-256 mappings hash, update the review sheet status line accordingly, and add new tests to pin the behavior.

File summaries

File	Description
src/lib/medication-badges.ts	Parse first unit strength robustly (decimals), and suppress badges for concentrations/combination products.
src/lib/medication-interactions.ts	Move documented low-severity tones off success; ensure incomplete-check verdicts degrade to neutral even from info.
scripts/build-medication-lexicon-report.ts	Compute a stable mappings hash, render status line from sign-off + current hash, and warn when sign-off is stale.
docs/medication-interaction-lexicon-review.md	Regenerated review sheet with “NOT current” status line when sign-off isn’t hash-bound.
tests/medication-badges.test.ts	Add pinned cases + corpus-wide invariant to ensure badged strengths are real unit-strength tokens in source text.
tests/medication-interactions.test.ts	Pin tone mapping for documented severities and protect manual-review headline behavior when incomplete.
tests/medication-low-severity-verdict.dom.test.tsx	DOM-level guard that a documented interaction never headlines as “No alert found”.
tests/medication-lexicon-report-signoff.test.ts	Unit tests for hash stability/changes, sign-off parsing, and committed sheet status line correctness.
docs/design-system/adoption-manifest.json	Include the new test in the design-system adoption manifest list.

Review details

• Files reviewed: 9/9 changed files
• Comments generated: 3
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.