Overlap regression tests, dead mic removal, CSS layering (plan items 3-5)

[BigSimmo]

Summary

Implements the agreed follow-up plan from the design QA session (items 3–5; items 1–2 were live-database/ops work described in Notes):

• Element-overlap regression tests (tests/ui-overlap.spec.ts, wired into the Playwright testMatch): asserts visible header elements never stack at 640–1536px and the composer clear button occupies its own slot. This bug class shipped three times (header ledger under the mode pill, clear button over typed text, standalone-home chips under the pill); the page-overflow smoke checks structurally cannot catch it.
• Remove non-functional voice input controls: the mic buttons in the composer, applications launcher, and document viewer had no click handler, and the forms search field drew a mic icon implying voice support that does not exist. Removed until the feature is real (plus unused imports and orphaned .answer-footer-search-mic CSS).
• CSS cascade layering (staged): audited every globals.css component class for same-property utility conflicts at call sites; the eight conflict-free helpers moved into @layer components so Tailwind utilities override them normally. The chrome classes stay intentionally unlayered (call sites rely on the class winning today) — documented with an in-file warning and a "CSS cascade layering" section in docs/process-hardening.md.
• Prettier pass over four drifted files (one from this branch's mic removal, three drifted on main).

Verification

• npm run verify:cheap — lint, typecheck, unit tests passed
• npm run verify:ui — 64 Chromium Playwright tests passed (includes the 10 new overlap tests)
• npm run format:check — green repo-wide after the prettier pass
• npm run verify:release — not run (no release claim)
• npm run check:supabase-project — Clinical KB Database (sjrfecxgysukkwxsowpy) confirmed (run as part of the migration work below)
• npm run check:deployment-readiness — not run (no deployment behavior changed)

Clinical Governance Preflight

Presentation/test/process changes only; no ingestion, answer generation, search/ranking, document access, or clinical output behavior touched in this diff.

• Source-backed claims still require linked source verification before clinical use
• No patient-identifiable document workflow was introduced or expanded
• Supabase target remains Clinical KB Database (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
• TGA SaMD impact — N/A, no decision-support behavior changed

Notes

• Plan item 1 (done, outside this diff): audit migration 20260702000000_commit_generation_preserve_legacy_artifacts applied to the live Clinical KB Database with explicit user approval; live function verified to carry the M13 guard; reindex:health green (2,065 docs indexed, 0 failed). check:indexing cannot run on this machine (pre-existing local Python/Tesseract prerequisite gap, already documented in process-hardening "Known limits").
• Plan item 2 (verified, nothing to do): all 41 audit findings are fixed or dispositioned on main via PR fix: remediate 2026-07 repo audit findings (clinical answer path, data integrity, privacy) #123 (spot-verified H1, H4, M10, M15 code paths). The one outstanding operational step — RAG_QUERY_HASH_SECRET — is now set in both local .env.local files (takes effect on next server restart); set it in any deployed environment that logs real queries.
• Five hybrid-fix migration files (20260701010000–20260701060000) are committed to the repo but absent from the live migrations table — consistent with the documented codify/reconcile flow (20260701140631 + the two 20260702 reconciliations are applied). Flagged for awareness, deliberately not applied.

🤖 Generated with Claude Code