[tyler-dane]

Fixes #

What and why

No issue: this is the weekly technical-debt sweep over merges from the last
seven days. Six behavior-preserving cleanups, all in code landed by #4763,
#4762, #4740, #4722 and #4741.

• Duplicated HTML escaper. apps/booking-web/render-meet-landing-html.ts
  hand-rolled two partial escapers (escapeHtmlAttr skipped >,
  escapeHtmlText skipped ") while a complete escapeHtml already existed
  in packages/backend. Moved that helper and its test to
  @core/util/escape-html.util (the package that owns the concept, reachable
  from both a frontend app and the backend) and pointed all four call sites at
  it. Output for the current landing copy is byte-identical; the copy contains
  no &, <, > or ".
• Copy-pasted meta stripping. Five near-identical html.replace(...) calls
  for superseded base-shell meta tags collapsed into one pattern list, which
  also lets html be a const.
• Dead timezone guard. autoEffectiveTimeZone returned
  browserTimeZoneStore.get() from both arms of its isUtcLikeTimeZone
  check, leftover scaffolding from fix(web): calendar zone consistency when browser reports UTC #4763 that read as if UTC hosts were
  special-cased. Dropped the branch and folded the function into
  currentEffectiveTimeZone, plus the two single-call pass-through wrappers
  applyDayjsDefault and getStoredBrowserTimeZone.
• Redundant welcome-step guard. emitWelcomeStepViewedOnce probed
  sessionStorage availability three times across two helpers and tested set
  membership twice, with the second test unreachable (readViewedSteps already
  returns a copy of the in-memory set when storage is unavailable). Now one
  probe and one check.
• Key lists duplicated as types. The event-form date field key unions were
  written out longhand and repeated as hand-maintained Set<string>
  literals, two places to keep in sync. The unions are now derived from the key
  lists, so switch exhaustiveness still covers exactly the accepted keys.
• Per-event timezone lookups. calendarIdsWithVisibleEventsOnDay called
  getEffectiveTimeZone() and re-formatted the view's civil day once per timed
  event. Both are now resolved once per scan.

No public behavior or contract changes; no test was weakened, and none was
added because none of this changes behavior.

Verify

Selected packages: core, web, backend, booking-web
Checks run: test:core, test:web, type-check, lint, knip, test:a11y
Checks skipped: (none)
Failed: test:backend:fast, test:booking-web, test:e2e
VERDICT: FAIL

All three local failures are this sandbox, and CI contradicts each one:

• test:backend:fast — 13 failures (GET /api/config, SSE Server) from
  missing local backend env, not an assertion: ZodError on BASEURL,
  FRONTEND_URL, MONGO_URI, SUPERTOKENS_URI, SUPERTOKENS_KEY, TZ. The
  identical 13 reproduce on a stashed clean tree at this PR's base, and
  unit-leg (backend) is green in CI. This PR's only backend edit is an import
  path.
• test:booking-web — passes standalone twice here (148 pass, 0 fail). It
  failed only inside verify, which runs it alongside the Playwright suite in
  a sandbox that cannot afford both. Green in CI.
• test:e2e — sandbox Playwright. In CI, shards 1, 2 and 4 passed; shard 3
  hit its 8-minute step cap at 64 of 67 specs passed with zero failures, which
  is a pre-existing budget problem on this shard rather than anything in this
  diff. Evidence and the base-commit timings are in
  this comment.

Green locally: test:core (1009 pass), test:web (714 pass),
test:booking-web (148 pass), type-check, lint, knip, test:a11y.
Green in CI on 25ac0db: unit and all unit-leg legs, static, delta,
boot-bundle (+0.0 KB), macos, e2e-shard 1/2/4, CodeQL.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TzeokDYpEzrNDZeYkA1DzW

[github-actions]

Boot bundle: 513.3 KB gzip (+0.0 KB vs main)

[tyler-dane]

e2e-shard (3, e2e/timed e2e/dst e2e/event-form e2e/parity) — step timeout, not a failure in this diff

The failing job ended with:

Running 67 tests using 1 worker
context canceled
································································
##[error]The action 'Run e2e tests' has timed out after 8 minutes.

64 dots, no F and no ✘: 64 of 67 specs passed and none failed. The job was
cut off by the step's timeout-minutes: 8, mid-run.

Why this isn't this PR's

Shard 3 already runs at ~75% of its budget on this PR's own base commit
(47703fb). From the base's E2E run, Run e2e tests per shard:

shard	specs	duration	budget
1	accessibility, allday	2m36s	8m
2	booking, oauth, auth, billing	2m13s	8m
3	timed, dst, event-form, parity	6m06s	8m
4	onboarding, calendars, life, navigation, attendees	3m09s	8m

A shard sitting at 6m06s is one slow runner away from the 8m cap on any PR.
.github/workflows/test-e2e.yml says these lists were balanced to put "each
shard near 90s" in the 2026-09 audit, so shard 3 has drifted about 4x past its
intended share, mostly from specs added to e2e/timed, e2e/dst and
e2e/parity over the last week.

This diff changes no spec and no test helper. It is a behavior-preserving
refactor that only removes work (one getEffectiveTimeZone() call and one
dayjs format per timed event become one per day-column scan), so it cannot
make a shard slower. Shards 1, 2 and 4 all passed here.

What I did and did not do

Re-running the failed job once (my single allowed re-run) to confirm the
timeout rather than a hidden failure.

I am not rebalancing the shard lists in this PR. That fix exists and is
straightforward, moving a directory from shard 3 to shard 1 or 2 and updating
packages/scripts/src/testing/e2e-shards.test.ts, but it is a CI-config change
unrelated to this refactor, and folding it in would widen the PR and put the
required e2e check's own balance under the same commit as the cleanup it is
meant to gate. It wants its own PR, and I have flagged it as the week's other
debt item.

---

Generated by Claude Code

[tyler-dane]

Re-run confirms it was the budget, not a hidden failure: shard 3 passed all 67
specs in 5m23s on this same commit 25ac0db, against the 7m56s cut-off on
attempt 1. e2e is green and the PR is in the merge queue.

Same commit, same specs, 5m23s vs >7m56s is runner-speed variance on a shard
that sits at 5-6 minutes of an 8-minute cap. Numbers for whoever picks up the
rebalance:

	attempt 1	attempt 2	base 47703fb
shard 3 Run e2e tests	timed out at 8m00s (64/67 passed, 0 failed)	5m23s	6m06s

---

Generated by Claude Code