Remove 63 dead duplicate dose records, and guard against regrowth

[monchee]

Two commits. The first is pre-existing work found uncommitted in the working tree; the second is the actual fix.

1. polish: dashboard and chrome cleanup

Work found complete and uncommitted in the working tree, not authored in this session. Committed separately so it stays distinct and is easy to revert if it was not meant to land.

UI pass across 15 files, net −121 lines: removes redundant dark: variants where the base class already resolves, adds focus-visible ring offsets, drops a backdrop-blur, applies whitespace-nowrap to the footer strapline. Removes the RecentTestingActivity dashboard component and its test; no references remain.

Verified before committing: tsc clean, full suite green at 98 files / 851 tests.

2. fix: remove 63 dead duplicate dose records

dreamOnlyProtocols.ts held 109 records but drugMasterlist.ts referenced only 50. The other 63 were dead copies of drugs since migrated to SCRATCH — Cefepime, Ceftriaxone, Cefotaxime, Alfentanil, Azithromycin, Chlorhexidine and more — left behind as each tranche moved a drug from findDreamOnly to findGenerated.

Nothing read them, so no test covered them and SCRATCH's clinical sign-off gate did not apply. They could drift from the source of truth silently, and someone correcting a dose could edit the dead copy, see no effect, and believe it fixed. That is exactly the duplication this project exists to eliminate, regrowing one tranche at a time — with 28 SCRATCH pages still to migrate, it would have kept growing.

The guard matters more than the deletion

A new test fails when dreamOnlyProtocols.ts contains an unreferenced record, and when drugMasterlist.ts references one that does not exist — both directions. Proven by re-adding a dead Cefepime entry: it failed and named the record. Without it, the problem simply returns.

Verification

Check	Result
Records	109 → 50, matching all 50 references
Unreferenced / referenced-missing	0 / 0
Guard fails on a dead record	✅ proven
Ordering	117 positions, 0 mismatches
tsc / lint / build	0 / 0 / 0
Tests	854 across 98 files
Coverage	78.8 / 70.9 / 72.8 / 80.0 vs 76 / 67 / 67 / 79

Ordering matters because DREAM selects protocol variants by array index — a positional shift would silently re-point a saved clinical plan at a different dose.

Note: 4 of the 50 references use the two-argument form findDreamOnly("Atropine", "skin") for drugs with an empty protocolLabel. Any tooling that parses these must handle both arity forms; a three-argument-only regex reports false orphans.

Left untracked deliberately

.github/agents/, .github/hooks/, .github/skills/ and four home-*.png screenshots — local tooling and visual-review artefacts. Probably worth gitignoring rather than committing.

🤖 Generated with Claude Code