[callumalpass]

Cause and smallest fix

Main run 41043845004 failed because the evening recurrence regression supplied an October 27 mutation timestamp but let the pinned model's recurrence helper read the real system date. Once October 29 passed, it advanced to a later occurrence. Freeze and restore the system clock in that test. No production code, model inputs, or expected behavior changed. No existing open PR covered the failure.

The requested full browser preflight crossed into Saturday and exposed three more tests assuming exactly four overdue demo tasks. The demo has five on Saturday. Those tests now measure the actual initial count and assert the exact move/completion/undo delta, rather than hard-coding a weekday.

Validation

Before opening:

• Full pnpm verify via rpnpm: PASS (235 files / 1,939 tests; release tests 394 pass / 1 skip; coverage and layer thresholds, i18n, format/typecheck/lint, both conformance checks, production build).
• Full pnpm test:e2e: 312 passed, 20 skipped, 2 failed (navigation-abort tracker and mobile heading lookup); both passed with --last-failed.
• Capacitor iOS and Android sync: PASS using the verified production build.
• Java 21 Android Gradle :app:testDebugUnitTest :app:lintDebug :app:assembleDebug: PASS. An initial Java 25 attempt was unsupported; the Java 21 rerun passed.
• Xcode compile unavailable on Linux; PR iOS CI supplies simulator compilation.

At current head c91b7cc:

• Android build, iOS simulator build, and production smoke PR CI: all SUCCESS.
• Six formerly weekday-dependent desktop/mobile tests: all PASS on Saturday.
• The complete local browser suite also passed twice on the SDK-cleanup branch built on this head: 314 passed / 20 existing skips each time (not claiming that as a separate run of this branch's older client graph).

Nothing merged; no workflow dispatch, store submission, or deployment performed.