## Outcome: **NOT READY for Sprint-5A acceptance**

### Important findings

1. **Missing days are removed instead of rendered as gaps.**  
   `dashboard-v5.js:289-308` constructs the chart axis only from non-null observations. A missing calendar day disappears, so `spanGaps: false` cannot prevent the line from connecting adjacent rendered points. This violates the no-interpolation requirement.

2. **Cockpit features are not implemented yet.**  
   `dashboard-v5.js:158-175` renders only value, date, and baseline value:
   - no sparkline,
   - no measurement age,
   - no baseline delta,
   - no coverage/quality badge.

3. **No main-trend fallback.**  
   `dashboard-v5.js:463-466` hardcodes `symptom.total`. If unavailable, the chart is empty instead of selecting the first usable registry metric and labeling the fallback.

4. **Event “lane” is only a badge list.**  
   `dashboard-v5.js:250-255` renders textual badges without a shared date axis. Events are separate from the main dataset, which is good, but this is not yet a temporally aligned event lane.

5. **Baseline registry contract is not honored.**  
   `data_provider.py:333-338` calls `rolling_baseline(points, metric.baseline_days)` and ignores `metric.baseline_min_observations`. Registry metrics specify a minimum of 3, while the helper defaults to 7.

6. **Series contract permits unsafe quality/value combinations and unordered data.**  
   `contracts.py:137-150` currently accepts:
   - numeric `value` with `quality="incomplete"`,
   - null `value` with `quality="complete"` or `"observed"`,
   - unsorted series dates,
   - future measurement dates.  
   The frontend assumes the final array element is latest, so unsorted valid bundles can display the wrong measurement.

7. **Concurrent render change introduced duplicate navigation.**  
   Current `render.py:44` and `render.py:67` both emit the same tablist. This risks duplicate accessible tabs and Playwright strict-locator failures.

### Exact minimal contract changes

No new serialized fields are required; keep `health_dashboard.bundle.v2`. Existing `today`, metric metadata, point `value/baseline/quality`, registry ordering, series, and events are sufficient.

Strengthen series validation only:

- dates must be strictly ascending and unique;
- measurement dates must be `<= bundle.today`;
- `quality == "incomplete"` requires `value is None`;
- `quality in {"complete", "observed"}` requires a finite numeric value;
- baseline must equal the median of prior non-null observations in `[date-baseline_days, date)`, once `baseline_min_observations` is reached; otherwise it must be null.

Treat metric registry order as the deterministic cockpit fallback order, with `symptom.total` first. No extra `fallback_metric_id`, summary, age, delta, coverage, or sparkline fields should be added because those are deterministic presentation derivatives.

### Minimal implementation/test set

- Pass `metric.baseline_min_observations` into `rolling_baseline`.
- Build a dense calendar axis for 7/30/90/180-day views; insert explicit `null` values and retain `spanGaps: false`.
- Derive measurement age from `bundle.today`, not wall-clock time.
- Show neutral signed absolute delta: `value - baseline`; do not label changes as better/worse, normal/abnormal, or safe/unsafe.
- Show factual coverage such as `6/30 dokumentierte Tage`, avoiding medically suggestive traffic-light thresholds.
- Render one sparkline per card using the dense axis.
- Select the first registry metric with usable period data for the main trend; expose its label and fallback context.
- Render events on a separate but date-aligned lane; never add them to the health-value dataset.
- Remove the duplicate navigation.

Required regressions:

- baseline starts exactly after the configured third prior observation and excludes current/future values;
- contract rejects unordered/future dates and inconsistent quality/value pairs;
- generated Chart.js datasets contain an explicit null for a wholly absent day and use `spanGaps: false`;
- six cards expose sparkline, deterministic age, baseline delta, and exact coverage;
- symptom-empty bundle falls back to the expected next registry metric;
- all-empty bundle retains the explicit empty state;
- an event on a measurement-gap day appears only in the event lane;
- visible copy contains no diagnosis, causality, reassurance, urgency, or treatment language;
- exactly one navigation tablist exists.

### Verification

- Focused Python suite: **16 passed**
- JavaScript syntax check: **passed**
- `git diff --check`: **passed**
- Browser suite not run because no test server URL was provided/started.
- No files were created or modified by this audit; the working tree already contained concurrent modifications to `data_provider.py`, `render.py`, tests, and the untracked Sprint-5A document.