# Quality gates

A release claim must have evidence. Use this list for the actual pages and states changed. Do not describe unperformed checks as passing. Automated generation validates data structure; it does not establish visual quality, accessible interaction or asset permission.

## Before browser review

- Confirm the reader task, real copy, destinations and attribution. Remove invented claims and unfinished placeholders.
- Run `node scripts/build-layout-catalog.mjs` if layout source changed. It checks 200 unique IDs and section sequences, known section types and desktop span totals.
- Inspect imports and file paths. Verify referenced images, audio and animation files exist and have appropriate provenance.
- Check HTML structure, heading outline, labels, button types and link semantics.
- Ensure form endpoints and visible states reflect real behavior.

## Browser matrix

| Context | Verify |
| --- | --- |
| 1440px and 1024px | Composition, line lengths, image crops, focus visibility |
| 768px and 390px | Reflow, source order, usable controls, no clipped labels |
| 320px | No page-wide horizontal overflow or lost essential information |
| 200% text enlargement / high zoom | Reflow, controls and fixed content do not overlap |
| Keyboard only | Skip link, every control, meaningful focus order, return paths |
| Touch | No essential hover dependency; controls and spacing usable |
| Reduced motion and explicit still | Complete content, no required travel or glyph animation |
| Sound off/on/preview | Clear control, one-preview semantics, visible equivalent states |
| Hidden tab / rapid input | No burst of stale audio or runaway motion |
| Missing assets / slow network | Content remains usable and placeholders are honest |
| JavaScript unavailable | Core content, navigation and non-enhanced actions remain usable |

Check at least the browsers in the consuming project's support policy. Run an automated accessibility scan and a manual screen-reader pass on important journeys; record the versions and scope. Review contrast for text, controls, focus and status against their actual backgrounds. Check both selected color variants and schemes if they are exposed to visitors.

## Interaction states

Review default, hover, focus, pressed/selected, disabled, pending, success, error, empty and offline states where relevant. A CSS hover demo does not verify a keyboard interaction. A syntactically valid form is not a successful server operation. A selected card is not saved until storage confirms it.

## Performance and content

Inspect layout shift and responsiveness during real use. Reserve image space, send appropriate sizes and avoid loading a player or audio context before needed. Check console errors and network failures. Use performance measurements only with recorded device, browser, network and test conditions; no universal performance score is implied.

## Completion record

```text
Page/revision:
Recipe ID and adaptations:
Automated commands and results:
Browsers/viewport sizes observed:
Keyboard/screen reader scope:
Motion/audio failure cases observed:
Content and asset rights reviewed:
Remaining limitations:
```

Keep this record brief but concrete. “Looks good” or “fully accessible” without scope is not an adequate validation claim.
