Correct clinical content errors found by the 20 Oct content audit
Held for owner clinical review — do not arm auto-merge or merge. Merge only after explicit owner approval (owner-seal hold; bots must not arm auto-merge).
Requested by Josh · project thread
Summary
Before: the 20 October 2026 clinical content audit found errors in medicine records, Mental Health Act forms and sections, services, calculators and dictionary entries. Six were high risk: the phenytoin albumin correction used the g/dL factor (about ten times too low with Australian g/L results), a buprenorphine pearl said a toddler "will likely just sleep", methadone induction allowed a first dose above the WA CPOP 25 mg limit, tranylcypromine showed a 60 mg/day maximum (PI 30), droperidol and ziprasidone used a 500 ms QTc cut-off (PI is lower), and flupentixol decanoate showed a UK 400 mg/week maximum labelled Australian.
After: every correction the audit recommended that did not need a new owner decision, new content authoring or a design mockup is applied, checked against the Australian PI or WA guideline the audit cited. Each edited item that was previously signed off is reset to awaiting clinical review, so nothing here claims Josh's sign-off. A new test pins the six high-risk corrections so a later import from the Medications repository cannot quietly bring the old wording back.
How: 629 checked edits to data/medications-snapshot.json (each edit verified the old value before replacing it), regenerated interaction index and lexicon report; forms 1A, 3D, 4B, 5A, 5B, 5C, 11D and 12C and Act sections 14 and 76 reset to drafted (section 14 now stores the Act's own text); three ARFID specifier rows reset to clinician review pending; a separate urgent-routing rule so a medical emergency points to 000; calculator wording, dictionary, First Nations, therapy and service corrections. Tests whose counts or fixtures assumed the reset items were signed are updated to the new counts; no test case was removed.
Areas touched: Clinical reference content, App experience, Prototypes, Data platform and access
RAG impact: no retrieval behaviour change. No ranking-protected file is edited. Medicine catalogue order for one broad query moved because clozapine's record now names its regular blood tests.
Verification
- Checks run:
- Full offline unit suite before the test expectation updates:
Tests 23 failed | 33004 passed, all 23 traced to the intended sign-off resets, the moved catalogue order and one retired source host. After the updates, the 13 affected files:Test Files 13 passed (13),Tests 257 passed (257) tests/medication-content-audit-corrections.test.ts: 6 passedcheck:clinical-hazard-controls,check:mha-act-sections,check:forms-review-sheet,check:calculator-content,check:specifiers-search-index,check:therapy-data-index,check:medication-interactions,check:medication-lexicon-report,check:source-catalogue: all passnpm run lintandnpm run typecheck: pass.check:diff-integrity:PASS — 18 changed test file(s), 340 -> 360 test case(s)- UI verification not run: the two UI files change text only (the copied calculator summary now carries the PHQ-9 item 9 flag, and one page description), with no layout change; covered by the calculator unit tests.
check:source-acquisitionsstill fails exactly as onmain(the lithium source, audit item R6-F-01, waits on Josh's decision)
- Full offline unit suite before the test expectation updates:
Risk and rollout
- Risk: clinical content. Keep this PR in draft. Josh reviews the content and records his own sign-off with
npm run clinical:review. Until he does, the reset forms, Act sections and specifier rows show as awaiting review. - Rollback: revert the merge commit.
- Provider or production effects: none from this branch. Merging deploys the corrected content to psychiatry.tools. No database change.
- The medicine records here are a copy of the Medications repository. The same corrections must go into that repository before the next import, or the new pinning test will fail the import.
- Two engineering drift exceptions are recorded in
docs/clinical-hazard-controls.json(expire 4 Dec 2026). They are not clinical approval. Hazard H8 needs Josh's reseal for the routing change.
Clinical Governance Preflight
- Source-backed claims still require linked source verification before clinical use
- No patient-identifiable document workflow was introduced or expanded without explicit governance approval
- Supabase target remains
PsychSift Production(sjrfecxgysukkwxsowpy) - Service-role keys and private document access remain server-only
- Demo/synthetic content remains clearly separated from real clinical sources
- Source metadata, review status, and outdated/unknown-source behavior remain conservative (every edited signed item is reset to awaiting review)
- Deployment classification/TGA SaMD impact was checked when clinical decision-support behavior changed (content corrections only; the routing change sends medical emergencies to 000, which is more conservative and makes no diagnosis or treatment decision)
Notes
- Items held for Josh's decision, his own words, a mockup or new content are listed in the audit review pack and in the project thread, not changed here.
- Report:
reviews/2026-10-20-clinical-content-accuracy.mdin the project files.
Generated by Claude Code
Summary by cubic
Corrects clinical content found by the 20 Oct audit across medicines, Mental Health Act forms and sections, services, calculators, and reference data. Items that were already signed off now await clinical review; this change does not imply clinical approval.
- Medical emergencies (seizures, collapse, unconsciousness, loss of consciousness, not breathing) pin 000 first, ahead of regional crisis routes, including when a regional place is named.
- Tests pin the six high-risk medication corrections to catch regressions from future imports.
Review notes
- Clinical owner review and
npm run clinical:reviewsign-off are required before the edited content is treated as approved. - Apply the medication corrections to the Medications repository before its next import.