[tyler-dane]

What and why

A debt pass over the last 7 days of merged changes (2026-10-15 to 2026-10-18).
No issue: this is the scheduled tech-debt sweep, not a product change. Two
behavior-preserving cleanups, one commit each.

1. refactor(web): drop dead date field keyboard flags

EventFormDateField grew two boolean props, arrowDownOpensCalendar and
keyboardOpensCalendar. Both defaulted to true, and no call site ever
passed false: DatePickers passed arrowDownOpensCalendar explicitly as
true, TimePickers relied on the default, and keyboardOpensCalendar was
never passed at all. The three branches they guarded were configuration
nobody could reach, and no test covered the false side. Both props are
gone; every caller keeps the behavior it already had.

While in that handler:

• switch (true) becomes a plain switch (event.key). Two of its cases
  carried a && !isOpen term that the if (isOpen) return above already
  guaranteed, so the ladder was hiding dead conditions.
• the repeated preventDefault() + stopPropagation() pair collapses into
  one consume helper.
• commitTypedDate is inlined. It had one caller and its boolean return was
  discarded.

2. refactor(web): share the duplicate-copy grouping pass

The holiday dedupe in #4553 added a second copy of the grouping loop in
mergeDuplicateCopies: walk data.ids, resolve the event and its calendar,
build a group key, bucket by it. The two copies had already drifted - only
the holiday pass skipped events an earlier pass dropped, and only the ical
pass guarded a missing event. Grouping now happens once behind a keyFor
callback, so each pass contributes only its own key rule. Both guards hold
for both passes, which is a no-op in practice: nothing is dropped before the
first pass runs, and that pass already required an icalUid that a missing
event could not supply.

Verify

Summary
Selected packages: web
Checks run: type-check, lint, knip, test:a11y
Checks skipped: (none)

Failed: test:web, test:e2e
VERDICT: FAIL

type-check, lint, knip and test:a11y (65 passed) are green. Both
failures reproduce on clean origin/main (5b0e9f8) in this sandbox with the
diff reverted, so neither is this change's:

• test:web - web shard 3/6 is killed at the runner's 5000 MB RSS rail
  (5198 MB here), stopping in
  apps/calendar-web/src/components/SelectView/SelectView.test.tsx. On
  origin/main the same shard trips the same rail at the same file
  (5015 MB). Shards 1 and 2 pass in full (654 and 760 tests). The memory is
  cumulative across the shard's 86 files, not one test: that file on its own
  passes 33/33 in 5s.
• test:e2e - 252 passed, 1 failed:
  e2e/calendars/calendar-experience.spec.ts:759 "a read-only event can be
  hidden from the menu, shown with x, and stays hidden after reload", on
  expect(hidden card).toBeVisible(). Re-run alone on origin/main it fails
  on the same locator with the same "element(s) not found".

Targeted runs covering the changed code are green:

• bun test:web -- apps/calendar-web/src/events - 397 pass, 0 fail
• bun test:web -- .../EventForm/DateControlsSection - 86 pass, 0 fail

The removed props are compile-time surface, so type-check covers their
removal repo-wide.

---

Generated by Claude Code

[tyler-dane]

Tech-debt sweep, 2026-10-15 to 2026-10-18

Window: the 50 commits merged to main in the last 7 days (99d9fda..5b0e9f8).
Hot spots by churn were the macOS Swift app and the web event form.

Addressed here (2 items, 1 PR):

#	Debt	Kind	Where it came from
1	arrowDownOpensCalendar / keyboardOpensCalendar were permanently true at every call site, with no test on the false side	dead configuration, premature generality	#4550, #4540
1b	switch (true) ladder hiding two && !isOpen terms the early return already guaranteed; duplicated preventDefault/stopPropagation; single-use helper with a discarded return	anti-pattern, repetition	#4541, #4550
2	Second copy of the copy-grouping loop in mergeDuplicateCopies, already drifted from the first (each had a guard the other lacked)	repetition, drift risk	#4553

Found and deliberately not acted on:

• NativeCalendarRootModel is 3,051 lines across 15 +Extension files and
  took 13 of the window's commits. That is the real structural debt in this
  window, but splitting a god object is a design call, and swift test is
  CI-only (no Xcode on a Linux sandbox), so a refactor would ship with no
  local signal. Worth a human-owned issue.
• TimePickers still carries the paired isStartMenuOpen / isEndMenuOpen
  booleans that refactor: collapse the all-day row-fit pair and banner dismissal state #4535 and refactor: share form jump ids and drop unused palette item fields #4543 collapsed for the date pickers. Collapsing
  them to one open side would make the two menus mutually exclusive, which is
  a behavior change, not a simplification.
• now-cue.store.test.ts has two tests whose names promise timer behavior
  ("clears after the pulse duration", "extends the active window") but assert
  only what resetForTests() does. The honest fix is to inject the scheduler
  into the store rather than stub globalThis.setTimeout, which is a wider
  change than a debt sweep should make unreviewed.

Pre-existing failures this sweep surfaced (both reproduce on clean
origin/main with the diff reverted, details in the PR body):

• bun test:web cannot finish in a 5 GB sandbox: shard 3/6 trips the
  runner's 5000 MB RSS rail partway through its 86 files.
• e2e/calendars/calendar-experience.spec.ts:759 ("a read-only event can be
  hidden from the menu, shown with x, and stays hidden after reload") fails on
  expect(hidden card).toBeVisible().

Labeled agent-automerge with auto-merge enabled; the merge queue and
agent-loop-merge-guard.sh decide. No merge has landed yet.

---

Generated by Claude Code

[github-actions]

Boot bundle: 509.8 KB gzip (+0.0 KB vs main)