perf(scripts): settle the alignment audit on DOM quiescence, not a flat budget by UncertaintyDeterminesYou4ndMe · Pull Request #2453 · apache/maka

UncertaintyDeterminesYou4ndMe · 2026-08-07T16:00:53Z

Part of #2389 (alignment audit; Pi TUI landed as #2448, Runtime Host as #2450, desktop e2e in its own PR).

What changed

The alignment audit slept a flat 2.5s per fixture after its readiness selector to cover content that populates after the shell mounts — 27.5s of unconditional waiting across 11 fixtures on every CI run, sized for the slowest runner and paid by every runner.

Post-mount content lands as DOM mutations, so the audit now waits for that observable fact: a MutationObserver quiet window (AUDIT_QUIET_MS, 500ms with no mutations), bounded by the unchanged 2.5s budget (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule.

Two properties keep coverage intact:

It runs afterwithFixtureWindow's own settle expression, which has already frozen animations/transitions and awaited document.fonts — so the only remaining movement is real content landing, and animation churn can't fake "busy".
The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget measured fully populated is still measured fully populated.

Timing (local, full audit)

	Before	After
Wall	42.5s	22.7s
Verdict	all 11 fixtures clean	all 11 fixtures clean

…at budget Part of apache#2389. The alignment audit slept a flat 2.5s per fixture after its readiness selector, to cover content that populates after the shell mounts (the QR-onboarding modal, the usage stats tables) — 27.5s of unconditional waiting across the 11 fixtures on every CI run. Content arriving after mount lands as DOM mutations, so the audit now waits for the observable fact instead: a MutationObserver quiet window (500ms with no mutations, AUDIT_QUIET_MS), bounded by the same 2.5s budget as before (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule. It runs after withFixtureWindow's own settle expression, which has already frozen animations and awaited document.fonts — the only remaining movement is real content landing. The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget covered is still measured fully populated. Timing (local, full audit, all 11 fixtures clean before and after): 42.5s -> 22.7s wall.

UncertaintyDeterminesYou4ndMe · 2026-08-07T16:00:53Z

Part of #2389 (alignment audit; Pi TUI landed as #2448, Runtime Host as #2450, desktop e2e in its own PR).

What changed

The alignment audit slept a flat 2.5s per fixture after its readiness selector to cover content that populates after the shell mounts — 27.5s of unconditional waiting across 11 fixtures on every CI run, sized for the slowest runner and paid by every runner.

Post-mount content lands as DOM mutations, so the audit now waits for that observable fact: a MutationObserver quiet window (AUDIT_QUIET_MS, 500ms with no mutations), bounded by the unchanged 2.5s budget (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule.

Two properties keep coverage intact:

It runs afterwithFixtureWindow's own settle expression, which has already frozen animations/transitions and awaited document.fonts — so the only remaining movement is real content landing, and animation churn can't fake "busy".
The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget measured fully populated is still measured fully populated.

Timing (local, full audit)

	Before	After
Wall	42.5s	22.7s
Verdict	all 11 fixtures clean	all 11 fixtures clean

…at budget Part of apache#2389. The alignment audit slept a flat 2.5s per fixture after its readiness selector, to cover content that populates after the shell mounts (the QR-onboarding modal, the usage stats tables) — 27.5s of unconditional waiting across the 11 fixtures on every CI run. Content arriving after mount lands as DOM mutations, so the audit now waits for the observable fact instead: a MutationObserver quiet window (500ms with no mutations, AUDIT_QUIET_MS), bounded by the same 2.5s budget as before (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule. It runs after withFixtureWindow's own settle expression, which has already frozen animations and awaited document.fonts — the only remaining movement is real content landing. The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget covered is still measured fully populated. Timing (local, full audit, all 11 fixtures clean before and after): 42.5s -> 22.7s wall.

UncertaintyDeterminesYou4ndMe · 2026-08-07T16:00:53Z

Part of #2389 (alignment audit; Pi TUI landed as #2448, Runtime Host as #2450, desktop e2e in its own PR).

What changed

The alignment audit slept a flat 2.5s per fixture after its readiness selector to cover content that populates after the shell mounts — 27.5s of unconditional waiting across 11 fixtures on every CI run, sized for the slowest runner and paid by every runner.

Post-mount content lands as DOM mutations, so the audit now waits for that observable fact: a MutationObserver quiet window (AUDIT_QUIET_MS, 500ms with no mutations), bounded by the unchanged 2.5s budget (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule.

Two properties keep coverage intact:

It runs afterwithFixtureWindow's own settle expression, which has already frozen animations/transitions and awaited document.fonts — so the only remaining movement is real content landing, and animation churn can't fake "busy".
The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget measured fully populated is still measured fully populated.

Timing (local, full audit)

	Before	After
Wall	42.5s	22.7s
Verdict	all 11 fixtures clean	all 11 fixtures clean

…at budget Part of apache#2389. The alignment audit slept a flat 2.5s per fixture after its readiness selector, to cover content that populates after the shell mounts (the QR-onboarding modal, the usage stats tables) — 27.5s of unconditional waiting across the 11 fixtures on every CI run. Content arriving after mount lands as DOM mutations, so the audit now waits for the observable fact instead: a MutationObserver quiet window (500ms with no mutations, AUDIT_QUIET_MS), bounded by the same 2.5s budget as before (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule. It runs after withFixtureWindow's own settle expression, which has already frozen animations and awaited document.fonts — the only remaining movement is real content landing. The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget covered is still measured fully populated. Timing (local, full audit, all 11 fixtures clean before and after): 42.5s -> 22.7s wall.

UncertaintyDeterminesYou4ndMe · 2026-08-07T16:00:53Z

Part of #2389 (alignment audit; Pi TUI landed as #2448, Runtime Host as #2450, desktop e2e in its own PR).

What changed

The alignment audit slept a flat 2.5s per fixture after its readiness selector to cover content that populates after the shell mounts — 27.5s of unconditional waiting across 11 fixtures on every CI run, sized for the slowest runner and paid by every runner.

Post-mount content lands as DOM mutations, so the audit now waits for that observable fact: a MutationObserver quiet window (AUDIT_QUIET_MS, 500ms with no mutations), bounded by the unchanged 2.5s budget (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule.

Two properties keep coverage intact:

It runs afterwithFixtureWindow's own settle expression, which has already frozen animations/transitions and awaited document.fonts — so the only remaining movement is real content landing, and animation churn can't fake "busy".
The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget measured fully populated is still measured fully populated.

Timing (local, full audit)

	Before	After
Wall	42.5s	22.7s
Verdict	all 11 fixtures clean	all 11 fixtures clean

…at budget Part of apache#2389. The alignment audit slept a flat 2.5s per fixture after its readiness selector, to cover content that populates after the shell mounts (the QR-onboarding modal, the usage stats tables) — 27.5s of unconditional waiting across the 11 fixtures on every CI run. Content arriving after mount lands as DOM mutations, so the audit now waits for the observable fact instead: a MutationObserver quiet window (500ms with no mutations, AUDIT_QUIET_MS), bounded by the same 2.5s budget as before (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule. It runs after withFixtureWindow's own settle expression, which has already frozen animations and awaited document.fonts — the only remaining movement is real content landing. The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget covered is still measured fully populated. Timing (local, full audit, all 11 fixtures clean before and after): 42.5s -> 22.7s wall.

UncertaintyDeterminesYou4ndMe · 2026-08-07T16:00:53Z

Part of #2389 (alignment audit; Pi TUI landed as #2448, Runtime Host as #2450, desktop e2e in its own PR).

What changed

The alignment audit slept a flat 2.5s per fixture after its readiness selector to cover content that populates after the shell mounts — 27.5s of unconditional waiting across 11 fixtures on every CI run, sized for the slowest runner and paid by every runner.

Post-mount content lands as DOM mutations, so the audit now waits for that observable fact: a MutationObserver quiet window (AUDIT_QUIET_MS, 500ms with no mutations), bounded by the unchanged 2.5s budget (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule.

Two properties keep coverage intact:

It runs afterwithFixtureWindow's own settle expression, which has already frozen animations/transitions and awaited document.fonts — so the only remaining movement is real content landing, and animation churn can't fake "busy".
The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget measured fully populated is still measured fully populated.

Timing (local, full audit)

	Before	After
Wall	42.5s	22.7s
Verdict	all 11 fixtures clean	all 11 fixtures clean

…at budget Part of apache#2389. The alignment audit slept a flat 2.5s per fixture after its readiness selector, to cover content that populates after the shell mounts (the QR-onboarding modal, the usage stats tables) — 27.5s of unconditional waiting across the 11 fixtures on every CI run. Content arriving after mount lands as DOM mutations, so the audit now waits for the observable fact instead: a MutationObserver quiet window (500ms with no mutations, AUDIT_QUIET_MS), bounded by the same 2.5s budget as before (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule. It runs after withFixtureWindow's own settle expression, which has already frozen animations and awaited document.fonts — the only remaining movement is real content landing. The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget covered is still measured fully populated. Timing (local, full audit, all 11 fixtures clean before and after): 42.5s -> 22.7s wall.

UncertaintyDeterminesYou4ndMe · 2026-08-07T16:00:53Z

Part of #2389 (alignment audit; Pi TUI landed as #2448, Runtime Host as #2450, desktop e2e in its own PR).

What changed

The alignment audit slept a flat 2.5s per fixture after its readiness selector to cover content that populates after the shell mounts — 27.5s of unconditional waiting across 11 fixtures on every CI run, sized for the slowest runner and paid by every runner.

Post-mount content lands as DOM mutations, so the audit now waits for that observable fact: a MutationObserver quiet window (AUDIT_QUIET_MS, 500ms with no mutations), bounded by the unchanged 2.5s budget (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule.

Two properties keep coverage intact:

It runs afterwithFixtureWindow's own settle expression, which has already frozen animations/transitions and awaited document.fonts — so the only remaining movement is real content landing, and animation churn can't fake "busy".
The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget measured fully populated is still measured fully populated.

Timing (local, full audit)

	Before	After
Wall	42.5s	22.7s
Verdict	all 11 fixtures clean	all 11 fixtures clean

…at budget Part of apache#2389. The alignment audit slept a flat 2.5s per fixture after its readiness selector, to cover content that populates after the shell mounts (the QR-onboarding modal, the usage stats tables) — 27.5s of unconditional waiting across the 11 fixtures on every CI run. Content arriving after mount lands as DOM mutations, so the audit now waits for the observable fact instead: a MutationObserver quiet window (500ms with no mutations, AUDIT_QUIET_MS), bounded by the same 2.5s budget as before (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule. It runs after withFixtureWindow's own settle expression, which has already frozen animations and awaited document.fonts — the only remaining movement is real content landing. The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget covered is still measured fully populated. Timing (local, full audit, all 11 fixtures clean before and after): 42.5s -> 22.7s wall.

UncertaintyDeterminesYou4ndMe · 2026-08-07T16:00:53Z

Part of #2389 (alignment audit; Pi TUI landed as #2448, Runtime Host as #2450, desktop e2e in its own PR).

What changed

The alignment audit slept a flat 2.5s per fixture after its readiness selector to cover content that populates after the shell mounts — 27.5s of unconditional waiting across 11 fixtures on every CI run, sized for the slowest runner and paid by every runner.

Post-mount content lands as DOM mutations, so the audit now waits for that observable fact: a MutationObserver quiet window (AUDIT_QUIET_MS, 500ms with no mutations), bounded by the unchanged 2.5s budget (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule.

Two properties keep coverage intact:

It runs afterwithFixtureWindow's own settle expression, which has already frozen animations/transitions and awaited document.fonts — so the only remaining movement is real content landing, and animation churn can't fake "busy".
The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget measured fully populated is still measured fully populated.

Timing (local, full audit)

	Before	After
Wall	42.5s	22.7s
Verdict	all 11 fixtures clean	all 11 fixtures clean

…at budget Part of apache#2389. The alignment audit slept a flat 2.5s per fixture after its readiness selector, to cover content that populates after the shell mounts (the QR-onboarding modal, the usage stats tables) — 27.5s of unconditional waiting across the 11 fixtures on every CI run. Content arriving after mount lands as DOM mutations, so the audit now waits for the observable fact instead: a MutationObserver quiet window (500ms with no mutations, AUDIT_QUIET_MS), bounded by the same 2.5s budget as before (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule. It runs after withFixtureWindow's own settle expression, which has already frozen animations and awaited document.fonts — the only remaining movement is real content landing. The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget covered is still measured fully populated. Timing (local, full audit, all 11 fixtures clean before and after): 42.5s -> 22.7s wall.

UncertaintyDeterminesYou4ndMe · 2026-08-07T16:00:53Z

Part of #2389 (alignment audit; Pi TUI landed as #2448, Runtime Host as #2450, desktop e2e in its own PR).

What changed

The alignment audit slept a flat 2.5s per fixture after its readiness selector to cover content that populates after the shell mounts — 27.5s of unconditional waiting across 11 fixtures on every CI run, sized for the slowest runner and paid by every runner.

Post-mount content lands as DOM mutations, so the audit now waits for that observable fact: a MutationObserver quiet window (AUDIT_QUIET_MS, 500ms with no mutations), bounded by the unchanged 2.5s budget (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule.

Two properties keep coverage intact:

It runs afterwithFixtureWindow's own settle expression, which has already frozen animations/transitions and awaited document.fonts — so the only remaining movement is real content landing, and animation churn can't fake "busy".
The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget measured fully populated is still measured fully populated.

Timing (local, full audit)

	Before	After
Wall	42.5s	22.7s
Verdict	all 11 fixtures clean	all 11 fixtures clean

…at budget Part of apache#2389. The alignment audit slept a flat 2.5s per fixture after its readiness selector, to cover content that populates after the shell mounts (the QR-onboarding modal, the usage stats tables) — 27.5s of unconditional waiting across the 11 fixtures on every CI run. Content arriving after mount lands as DOM mutations, so the audit now waits for the observable fact instead: a MutationObserver quiet window (500ms with no mutations, AUDIT_QUIET_MS), bounded by the same 2.5s budget as before (AUDIT_SETTLE_MS) so a never-quiet page still proceeds on the old CI-gate schedule. It runs after withFixtureWindow's own settle expression, which has already frozen animations and awaited document.fonts — the only remaining movement is real content landing. The quiet window outlasts the longest fixture-internal deferral (the settings-bots-onboarding waiting state arms ~300ms after mount), so every surface the flat budget covered is still measured fully populated. Timing (local, full audit, all 11 fixtures clean before and after): 42.5s -> 22.7s wall.

UncertaintyDeterminesYou4ndMe mentioned this pull request Aug 7, 2026
perf(test): replace fixed waits with observable completion #2389
Closed