## Sprint 5A synthetic acceptance audit and test plan

### Current state

- **Existing tests pass on the current synthetic fixture:**
  - `pytest tests/test_health_dashboard_v5.py`: **15 passed**
  - Playwright Chromium suite: **9 passed**
- Current suite covers:
  - Four viewports: `390×844`, `768×1024`, `1024×768`, `1440×900`
  - One visible panel, horizontal overflow, 44 px controls
  - Basic navigation, transient privacy, doctor print content
  - Reduced motion, basic 200% text reflow
  - Skip link and modal focus trap
  - Explorer limits and period-data behavior
  - Several empty chart states
- **Missing required viewport:** `1920×1080`
- **No screenshot assertions or captured acceptance screenshots exist.**

### Material Sprint 5A gaps

Runtime probes across all five target viewports found no horizontal page overflow and exactly one visible panel, but the current UI is not yet Sprint-5A complete:

- Desktop navigation remains a sticky/flex navigation placed after the main content, rather than an asserted desktop horizontal/sidebar mode.
- Mobile nav becomes fixed at `≤900px`, but tests do not verify desktop/mobile mode switching or content occlusion.
- Six Cockpit cards render, but:
  - **0 sparklines**
  - no visible coverage/quality status
  - date exists, but no explicit measurement age
- Cockpit chart is hardcoded to `symptom.total`; no fallback-metric behavior is implemented or tested.
- Event track is separate in the DOM, but there is no behavioral test proving events never become a value-series dataset.
- `Heute` tests only confirm its heading:
  - no tests for useful content without a complete symptom day
  - no source freshness
  - no short trend
  - no maximum-two hints
  - mobile label is “Symptome erfassen,” not the requested central `+ Erfassen`
- Fixture includes stale/future rows and partial symptom/nutrition data, but browser tests do not expose/assert the required current/stale/future states, low coverage, visible sparkline gaps, or fallback chart.

### Proposed deterministic Playwright suite

Use projects for exactly:

```text
desktop-xl   1920×1080
desktop      1440×900
tablet-land  1024×768
tablet-port  768×1024
phone        390×844
```

Pin locale/timezone, anchor date `2026-06-15`, reduced animation, device scale factor, and Chromium version. Start the real local server against a newly generated temporary fixture; hash the database before and after.

#### Core acceptance tests

1. **Shell matrix**
   - Open each primary view and More.
   - Assert exactly one visible `[data-dashboard-panel]`.
   - Assert `scrollWidth <= clientWidth` after every view.
   - Check console errors, page errors, failed requests, and non-local requests.
   - Measure all semantic controls, including nested inputs, at ≥44×44 px.

2. **Navigation modes**
   - At `1024`, `1440`, and `1920`, assert desktop navigation is visible as the designed sidebar/topbar and mobile fixed nav is absent.
   - At `768` and `390`, assert mobile nav is fixed and desktop nav is absent.
   - Confirm fixed nav does not cover the last focusable/content element.
   - Verify active state, roving tabindex, Arrow keys, Home/End, and `Mehr` `aria-expanded` reset.

3. **Today usefulness**
   - Use a fixture with no complete symptom day but available medication, source, and registry history.
   - Assert useful snapshot still renders.
   - Assert ≤3 prioritized tasks, separate administered/planned medication, compact freshness, short history, and ≤2 distinct hints.
   - On phone, assert a prominent `+ Erfassen`; on desktop, assert it does not displace primary information.

4. **Cockpit cards**
   - Assert exactly six core cards.
   - For each card, inspect value/unit, measurement date/age, baseline copy, coverage/quality text, and sparkline.
   - Inspect chart data, not labels alone:
     - null dates remain `null`
     - explicit zero remains `0`
     - `spanGaps === false`
   - Low-coverage fixture must show low/partial status and never present a partial symptom day as complete.

5. **Main-chart fallback and event lane**
   - Fixture A: complete symptom total available → selected chart metric is `symptom.total`.
   - Fixture B: no complete symptom total, registry metric available → deterministic documented fallback selected and clearly labeled.
   - Fixture C: no eligible metric → explicit empty state.
   - Assert event lane is a separate named semantic region and no event label/type appears in chart datasets.
   - Period changes must alter labels/data counts and event-lane membership.

6. **Screenshots**
   - Synthetic-only full-page screenshots for Today and Cockpit at all five viewports.
   - Additional phone navigation-open/check-in and desktop Doctor/print screenshots.
   - Store under an explicitly ignored artifact directory.
   - Stabilize fonts, animations, timezone, canvas rendering, and mask the per-request CSRF token if visible.

7. **Privacy**
   - Enable privacy in every data-bearing view.
   - Enumerate known sensitive text nodes and canvases; verify computed masking/blur, not merely a root class.
   - Confirm navigation remains operable, toggle restores content, and reload resets privacy.
   - Assert URL, storage, console, trace metadata, and network requests contain no fixture health values.

8. **Keyboard/accessibility**
   - Skip link, complete tab order, visible focus style, roving tabs, all period controls, More, details/summary, dialog open/close, Escape, Shift+Tab wrapping, and restored focus.
   - Check landmarks, heading order, tab/panel relationships, accessible chart summaries, status text independent of color.
   - Add automated accessibility scanning, while retaining manual keyboard assertions.

9. **Reduced motion and 200% text**
   - Run all views under reduced motion; inspect transitions/animations and chart animation configuration.
   - Test actual browser/CDP text zoom or equivalent 200% text scaling, not only one root CSS declaration.
   - Assert no horizontal overflow, clipping, overlap, unreachable controls, or fixed-nav occlusion.

10. **Empty/long-label/future anomaly profiles**
    - Schema-only empty fixture: meaningful state in Today, Cockpit, event lane, Explorer, Nutrition, Doctor, and More.
    - Long-label fixture: medication, food, lab, metric, and event labels at all narrow viewports.
    - Future-dated observation: visibly “future/check” and never selected as ordinary latest/fresh.
    - Current/stale/future states must be textually distinct and describe data quality, not medical risk.

### Important test pitfalls

- Existing 200% test only modifies `documentElement.style.fontSize`; this is weaker than browser text zoom.
- Existing privacy test can pass by checking only the first blurred panel child while nested content or canvases leak.
- Existing reduced-motion scan runs only the initially visible view.
- Existing period test mostly checks caption changes; only Explorer checks point counts.
- Empty-state interception mutates HTML after generation and can bypass backend contract validation; prefer real validated fixture variants.
- Screenshot/report directories are not explicitly ignored, although image/ZIP extensions happen to be broadly ignored.
- `CHROMIUM_PATH=/usr/bin/chromium` is environment-specific, and the package script does not start the fixture/server itself; clean-checkout browser execution is not currently one-command reproducible.
- Visual baselines must never be refreshed from a non-synthetic server.
- Avoid asserting CSS declarations as proof of runtime target size, privacy, nav mode, or reflow—measure computed browser state.

### Repository impact and issues

- **No repository files created or modified.**
- Temporary fixture, server, browser dependencies, screenshots, and test artifacts were removed.
- Concurrent uncommitted changes appeared during the audit in:
  - `scripts/health/dashboard_v5/data_provider.py`
  - `scripts/health/dashboard_v5/render.py`
  - `tests/test_health_dashboard_v5.py`
- The Sprint 5A document is currently untracked.