# CHANGES REQUIRED

## Important findings

### 1. Lag/phase panels ignore empty selection and global period

**Anchors:** `scripts/health/assets/health-assets/dashboard-v5.js:629-632`, `601-617`

The filter treats an empty metric selection as “show every correlation”:

```js
row.status === "computed"
&& (!correlationIds.size || ...)
```

Therefore, **Auswahl leeren** empties the trend but expands Lag and Phase to all validated results. Global period changes also alter the trend without changing or qualifying the aggregate scope.

**Exact synthetic reproducer:**

1. Generate the checked-in synthetic fixture and v5 dashboard.
2. Open Explorer → **Schlaf und Erholung**.
3. Click **Auswahl leeren**.
4. Inspect:

```js
({
  selected: [...document.querySelectorAll("#metric-picker input:checked")].map(x => x.value),
  trendEmpty: !document.querySelector("#explorer-chart-empty").hidden,
  lagCells: document.querySelectorAll("#explorer-lag-heatmap .explorer-result-cell").length,
  phaseCells: document.querySelectorAll("#explorer-phase .explorer-result-cell").length
})
```

**Observed:**

```json
{
  "selected": [],
  "trendEmpty": true,
  "lagCells": 30,
  "phaseCells": 30
}
```

Period probe, using **7 T** and then **Alle**:

```json
{
  "seven": {"trendPoints": "6", "lagCount": 15, "phaseCount": 15},
  "all": {"trendPoints": "19", "lagCount": 15, "phaseCount": 15},
  "lagIdentical": true,
  "phaseIdentical": true
}
```

**Consequence:** Lag/phase evidence appears associated with the current empty selection and visible period even though it represents a different server-side aggregate scope. Since browser recomputation is forbidden, these panels should fail closed for empty/unmatched selections and explicitly disclose—or server-contract—an aggregate period/cadence independent of the global controls.

---

### 2. Weekly trend and event lane do not use the same visible domain

**Anchors:** `dashboard-v5.js:537-540`, `613-615`, `405-424`

The weekly trend domain is constructed from Monday bucket keys, while the Explorer event lane always derives its domain from raw daily measurement dates.

**Exact synthetic reproducer:**

1. Explorer → **Schlaf und Erholung** → **Wöchentlich**.
2. Inspect:

```js
const canvas = document.querySelector("#explorer-chart");
const marker = document.querySelector(
  '#explorer-event-track .event-marker[data-event-date="2026-06-14"]'
);
({
  chartFirst: canvas.dataset.firstDay,
  chartLast: canvas.dataset.lastDay,
  eventDate: marker.dataset.eventDate,
  eventPosition: marker.style.getPropertyValue("--event-position"),
  eventAxis: document
    .querySelector("#explorer-event-track .event-axis")
    .getAttribute("aria-label")
})
```

**Observed:**

```json
{
  "chartFirst": "2026-05-18",
  "chartLast": "2026-06-08",
  "eventDate": "2026-06-14",
  "eventPosition": "100%",
  "eventAxis": "Ereignisse von 24.5.2026 bis 14.6.2026"
}
```

The existing browser test at `tests/browser/dashboard_v5.spec.js:240-276` currently asserts the `100%` event position but does not compare the event-lane domain with `data-first-day`/`data-last-day`.

**Consequence:** Event markers are visually positioned against a different temporal axis from the weekly chart. Use one explicitly shared domain—potentially week intervals if events retain exact daily dates—and test both endpoints.

---

## Other reviewed semantics

No additional blocker reproduced for:

- Missing daily observations remaining unknown
- Partial weeks remaining `null`
- Seven finite numeric days required for a weekly value
- Numeric zero preservation
- Registry-controlled raw `sum`/`avg`
- Index mode averaging daily `ratio_100` values
- Same-cadence complete-case scatter pairing
- Canonical registry/preset drift rejection
- Affirmative administered-event allowlist
- Aggregate-only correlation payload validation
- Non-causal/descriptive wording
- No persistence of scatter pairs or dates

## Verification

- `129 passed` — full Python suite
- `16 passed` — full Playwright suite against synthetic fixture
- Focused v5 tests: `22 passed`
- JavaScript syntax, Python compilation, and `git diff --check`: passed
- No production data accessed; no repository files modified
- Temporary synthetic fixture/server artifacts removed; final working-tree status matches the initial Sprint-5B tree