## Review result

**P0: 0**

### P1: 7

1. **Canonical rows can be marked original-verified without an original/content-review gate.**  
   `scripts/health/health_dashboard_action_worker.py:1754-1785`  
   Transfer requires candidate state/revisions but never requires `original_reviewed_at`, reviewed pages, or `content_status='reviewed'`. It then writes `validierungsstatus='validiert'` and `verified_against_original=1`.  
   **Counterexample:** synthetic execution transferred OCR CRP while document state was `(original_reviewed_at=None, content_status='partial')`; resulting canonical row was `('<4.2', ..., 'validiert', 1)`. This is operator-triggered but falsely asserts verification against the original.

2. **Document date is promoted into both specimen and report dates.**  
   `scripts/health/dashboard_v5/document_reconciliation.py:184-195`  
   `scripts/health/health_dashboard_action_worker.py:1781-1783`  
   Reconciliation uses `d.document_date` as the laboratory observation day; transfer writes that same day to both `abnahme_datum` and `befund_datum`.  
   **Counterexample:** a report dated June 10 with no documented collection date becomes “sampled June 10” and “reported June 10,” inventing two provenance facts.

3. **The advertised preview hash is accepted but never enforced.**  
   `scripts/health/health_dashboard_action_worker.py:449-458`  
   `scripts/health/health_dashboard_action_worker.py:1832-1840`  
   Only `expected_candidate_revision` is checked; `preview_revision` is format-validated but never recomputed or compared.  
   **Counterexample:** after the displayed preview, changing `normalized_date`, `reference_text`, and `comparison_digest` without changing `candidate_revision` was accepted; the stage used the changed date/reference rather than the operator-reviewed preview.

4. **Reference-range parsing can corrupt canonical bounds and ignores unit consistency.**  
   `scripts/health/health_dashboard_action_worker.py:1735-1741`  
   `scripts/health/health_dashboard_action_worker.py:1781-1783`  
   **Counterexamples:** `0-5 mg/L` is parsed as lower `0`, upper `-5`; `>5` loses the lower bound; a candidate in `mg/L` with reference text `0–5 mg/dL` is accepted without detecting the unit conflict. The raw reference text is not retained in the canonical row.

5. **Qualitative laboratory results are modeled but cannot complete the confirmation path.**  
   `scripts/health/dashboard_v5/lab_review.py:123-150`  
   `scripts/health/assets/health-assets/dashboard-v5-record.js:633-637`  
   `scripts/health/health_dashboard_action_worker.py:1765-1769`  
   Qualitative results without units are classified incomplete; even with a unit, the UI enables preview only for `kind==='numeric'`, and transfer requires a numeric substring.  
   **Counterexample:** a documented `ANA: negativ` with no unit can only be deferred/rejected/marked incomplete, never preserved as a confirmed qualitative observation.

6. **Censored/operator values become exact numeric points outside the history display.**  
   `scripts/health/dashboard_v5/read_api.py:652-681`  
   `scripts/health/dashboard_v5/read_api.py:3650-3665`  
   `scripts/health/assets/health-assets/dashboard-v5-api-explorer.js:224-248`  
   `scripts/health/assets/health-assets/dashboard-v5-record.js:791-802`  
   The API separates `<` from numeric `4.2`, but computes numeric deltas without checking the operator; Explorer tables/charts and doctor-report charts consume only `point.value`.  
   **Counterexample:** `<4.2 mg/L` is plotted and tabulated as `4.2`, can receive an exact delta/reference comparison, and loses `<` in the doctor chart.

7. **Duplicate handling can collapse distinct source-specific observations.**  
   `scripts/health/health_dashboard_action_worker.py:1771-1779`  
   An exact same-day/value/unit match reuses the existing canonical row ID and does not create a canonical provenance link for the second document or its observation-specific reference.  
   **Counterexample:** two distinct reports documenting the same CRP value but different source documents/reference ranges become one canonical observation; the second provenance survives only in review staging/audit data despite the UI claim of “Keine stille Zusammenführung.”

### P2: 1

1. **Thematic grouping is inferred by hard-coded medical-name heuristics, not supplied by the catalog.**  
   `scripts/health/dashboard_v5/lab_review.py:64-92`  
   `scripts/health/dashboard_v5/lab_review.py:166-197`  
   `LAB_CATALOG_SPECS` contains no reviewed thematic group, yet `_group()` infers categories from parameter/metric strings.  
   **Counterexample:** Ferritin is assigned “Vitamine und Mineralstoffe” solely because its name contains `ferritin`; changing an identifier or alias can silently change grouping without a catalog review.

## Verification

- Exact base/HEAD: `96df2a255dcfb73ea56257997a15e64d58824d64`.
- Reviewed all tracked changes and five nonignored untracked source/test/doc files.
- Focused synthetic tests: **11 passed**.
- Isolated Sprint 7C-E browser suite: **5 passed**.
- Python AST, JavaScript syntax, and `git diff --check`: passed.
- Used only isolated synthetic fixtures and source-controlled material; no production records were opened or exposed.
- **Files created or modified by review: none.**