[morsela]

Six places where the same code existed two to seven times over, each kept in step by hand. No behaviour change; the 451 existing tests pass unchanged and 15 new ones cover the wrappers this adds.

These came out of a duplication audit of the whole app. The remaining findings from that audit are not in this PR — see the end.

What was duplicated, and why each one mattered

Notification teardown, four times in drivers.ts. classic, fitshow and ks1234 each carried a byte-identical detach block — remove the listener, stopNotifications() in a try/catch, null the pair — and FTMS a fourth spelling of it as a loop. Four places to fix a teardown that ever turns out to leak, on the path that decides whether a pad is left bound to a tab somebody has walked away from. Now subscribe/unsubscribe. The event name is a constant too: it appeared as a string literal twelve times, and a typo in one is a listener that registers and never fires.

The GATT fake reimplemented two helpers it could have imported. It had its own toHex, duplicating the hex the package already exports, and its own fullUuid, duplicating a canonicalUuid that was module-private. That second one is the sharp edge: a fake whose UUID canonicalisation is a separate implementation can agree with itself and disagree with detectDriver, which is the one thing a fake must not do. canonicalUuid is now exported and the fake imports both. The classic suite had a third hex formatter of its own; it shares the same one now while still deriving the frame CRC independently, since an expectation taken from the code under test checks nothing.

isObj declared three times, identically, in settings.ts, session.ts and backup.ts — the last of which already imports both sanitisers from the other two and then kept its own copy of the primitive underneath them. Moved to lib/validate.ts, with the array case pinned by a test because that case is the reason the guard is written as it is: typeof [] === 'object', so the obvious spelling waves an array through and every field read off it is then silently undefined.

Storage keys declared in three modules and listed again by hand in Recovery.tsx. That screen needs the whole set to dump and to clear, so it held a second copy as string literals. A fifth stored key added anywhere would have dropped silently out of both the rescue download and the clear button — the last two controls that still work once everything else has failed. lib/storage.ts holds the keys once and derives the full list from them.

The same try/catch around localStorage, seven times. Every one was needed — the accessors throw outright with site data blocked, and setItem throws on quota in private mode — and the answer was always the same. Now four wrappers.

[...sessions.value, ...(currentSession.value ? [currentSession.value] : [])], six times, twice in components. That put "what counts as a session" in the view layer, where changing it would have had to be chased out of the store. One computed now, which also memoises the concatenation: streak and bestStreak each walk dailySeries(400), and the streak chain asks for both on every render.

Two judgement calls worth a look

readJson distinguishes undefined from a stored null. Missing, unreachable and unparseable all come back undefined; a stored null comes back as null. That is not tidiness — restoreOpenSession returns quietly when there is nothing to recover, but hands a value the app actually wrote to sanitizeSession so it gets rejected and reported like any other bad record. Collapsing the two would have turned one of those into silence.

The blanket catches in loadSessions and restoreOpenSession are kept. I removed them at first and put them back: they are not about storage any more, they are about running at module scope. main.tsx calls both before render(), so a throw there never reaches the error boundary and leaves a blank page instead of the Recovery screen — the exact failure this app went to some trouble to make impossible. They now say so in a comment.

Verification

npm run check --workspaces, npm test, npm run lint and npm run build all clean. 451 existing tests pass unchanged; 15 added, covering the storage wrappers' failure behaviour under a localStorage that throws — which is what those seven try/catch blocks existed for and what nothing previously tested — plus the key registry and the isObj array case. The four protocol suites passed unedited on either side of the driver changes, which is the real safety net there since there is no treadmill in CI.

Not in this PR

The audit also turned up: fmtMph existing while ten call sites hand-roll toMph(x).toFixed(1); the resume confirmation dialog written out twice; a copy-to-clipboard-then-flash pattern four times; the unsupported-browser fallback block twice; shared chart scaffolding between Area and Column (already drifted — axis labels sit at H - 3 in one and H - 4 in the other); four hand-rolled poll-until-deadline loops in connection.ts; duplicated test fixtures (fakePad, the session() builder); ~25 lines of boilerplate across the four static pages; and the model/protocol table in five copies, which has already drifted three separate ways. Happy to take any of those next.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TXUkCfJBbZiFPoJLoaiKBo

---

Generated by Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
beltcontrol	Ready	Preview	Sep 17, 2026 1:31am UTC