## Audit outcome

**Important — correction chains are double-counted outside the medication audit history.**

A synthetic 3-row chain (`administered → corrected/missed → corrected/administered`) produced:

- `/day`: three raw entries across three dates; the superseded origin remains administered.
- `/calendar`: medication count `1` on all three dates.
- `/events` and doctor-report `timeline`: three events.
- Explorer: stale origin is grouped as **Medikamente**, while correction events fall into **Sonstige** because `eventGroup()` recognizes only `medication_administered`, not `type=medication`.
- Medication history/doctor medication section correctly retains all three audit rows and marks:
  - origin `superseded_by_correction=true`
  - first correction `is_latest_effective=false`
  - terminal correction `is_latest_effective=true`

### Exact code risks

- `read_api.py:992-1042` — `_events()` queries raw rows; no correction supersession.
- `read_api.py:1360-1404` — `_day_medications()` queries raw rows and has no chain projection.
- `read_api.py:2698-2712` — calendar counts raw row IDs, including every correction revision.
- `read_api.py:3937-3959` — record overview’s “latest administered” can select a superseded origin and ignores an effective administered correction because it remains in the `corrected` bucket.
- `read_api.py:4301` — doctor timeline inherits raw `_events()` duplication.
- `dashboard-v5-api-explorer.js:252-262` — corrections are misclassified as `other`; stale administered origins remain medication markers.
- `dashboard-v5-record.js:961` — report chart markers use every raw `administered` item, including superseded origins, and omit terminal corrections whose effective status is administered.
- Doctor medication history itself is audit-correct, but its generic documented-row count at `read_api.py:4427-4430` counts all versions; it must not be interpreted as effective administration count.

**Important — day projection drops structured medication semantics.**

`_day_medications()` selects only legacy `dose` and `route`. A structured row containing:

- `actual_quantity_value="1"`
- `actual_dosage_form="Spritze"`
- `actual_strength="40 mg"`
- `route_original="subkutan"`
- `route_normalized="subcutaneous"`
- `injection_side="right"`

was returned by `/day` with empty `dose` and empty `route`. The current passing F1 test hides this because the worker also populates compatibility fields.

**Moderate — displayed enums are not localized in history details.**

`dashboard-v5-record.js:885` renders raw values:

- `item.status` → `administered`
- `item.route_normalized` → `subcutaneous`
- `item.injection_side` → `right`

The select controls are localized correctly (`Subkutan`, `Rechts`), but read-only details are not.

**Moderate — `route_original` fallback is only partially safe.**

`read_api.py:3641` performs fallback before validation:

```python
safe_metadata_text(row["route_original"] or row["route"], ...)
```

If `route_original` is non-empty but rejected as unsafe while legacy `route` is valid, the valid fallback is lost. Sanitize independently, then fallback:

```python
safe(route_original) or safe(route)
```

### Minimal implementation shape

Add one shared medication chain projection in `medication_contract.py`:

- annotate every row with `chain_origin_id`, `superseded_by_correction`, `is_latest_effective`;
- retain all rows for `/medications` and doctor audit history;
- expose terminal effective rows/statuses separately;
- make day, calendar, events/timeline, record-summary KPIs, and report chart markers consume only terminal effective rows.

Keep correction-record date versus origin-event date an explicit contract decision; do not derive it differently per endpoint.

Add shared renderer maps for status, route, and side enums, used by history details and report rendering.

### Minimal regression tests

1. **Single synthetic three-row chain**
   - `/medications` and doctor medication history retain all 3 rows with exactly one `is_latest_effective`.
   - Effective day/calendar/events/timeline surfaces count exactly 1 semantic event, not 3.
   - Effective status is `administered`.
   - Record-summary “latest administered” and report marker use the terminal correction, never the superseded origin.

2. **Structured projection**
   - Insert a structured event without compatibility `dose`/`route`.
   - Assert `/day` renders `1 Spritze · 40 mg`, with route fallback/localization.

3. **Renderer localization**
   - Browser assertion that details contain `Subkutan` and `Rechts`.
   - Assert raw `subcutaneous` and `right` are absent from visible detail text.

4. **Fallback counterexample**
   - Unsafe/non-displayable `route_original` plus valid legacy `route="subkutan"`.
   - Assert API returns the valid legacy fallback without exposing the rejected value.

### Verification

- Existing focused suite: **21 passed**  
  `pytest -q -p no:cacheprovider tests/test_dashboard_v5_sprint7c_f.py tests/test_dashboard_v5_sprint7c_f1.py`
- Repository remained clean at `a730ca4f0f3cf12c74dbb3e1c512ad1978c96fe3`.
- **Files modified:** none.