fix(results): gate operational AI exports on QA

docs/API_CONTRACTS.md

@@ -2220,10 +2220,34 @@ Detection run export request:
 {
   "export_kind": "detection_run",
   "analysis_run_id": "uuid",
+  "intended_use": "review",
   "name": "optional-basename"
 }
 ```
 
+`intended_use` is `review` by default. Every detection FeatureCollection and
+export record contains a machine-readable `geointel_result` / `result_trust`
+contract. Raw model output remains `authoritative: false` and is classified as
+`unverified_ai_review_output` when authoritative QA is absent or incomplete.
+Feature properties repeat the classification so it survives GIS workflows
+that discard collection-level metadata.
+
+`intended_use: operational` fails closed with
+`DETECTION_OPERATIONAL_EXPORT_BLOCKED` unless the same persisted run has a
+completed detection-versus-reference check that proves all of the following:
+
+- the reference Dataset is authoritative and specifically approved as primary
+  authority for building validation;
+- inference coverage was derived from the persisted tile-manifest union;
+- imagery/reference time compatibility is proven;
+- geometry is supported and no QA warnings remain;
+- false-positive and false-negative counts are both explicitly present and
+  zero.
+
+Even a passing operational export remains AI-derived and therefore retains
+`authoritative: false`, the exact quality-check/reference identifiers and an
+operator-review limitation. Model confidence by itself never unlocks export.
+
 Segmentation run export request:
 
 ```json
@@ -2251,7 +2275,10 @@ Response persists an `exports` row and writes a deterministic JSON artifact:
 
 Vector dataset exports use the stored dataset GeoJSON. Detection and
 segmentation exports use persisted first-class geometry records and the
-existing Detection/Segmentation GeoJSON conversion services. Vector selection
+existing Detection/Segmentation GeoJSON conversion services. Detection export
+in the frontend is deliberately labelled as a control layer and requests
+`intended_use: review`; it cannot silently produce an operationally approved
+artifact. Vector selection
 exports query persisted PostGIS `vector_features` with the supplied EPSG:4326
 bbox, write the selected FeatureCollection as a `vector_selection_geojson`
 artifact, and persist bbox/feature-count metadata in the export record. When

docs/CODEX_EXECUTION_LOG.md

@@ -12472,3 +12472,29 @@ Open:
   product benchmark remain required before any production-accuracy claim.
 - Every deployed model asset needs its own generated scope manifest and exact
   configured manifest checksum before enforced inference is available.
+
+## 2026-08-09 - Detection result trust and operational export gate
+
+### Changed
+
+- Detection GeoJSON is now explicitly a review/control layer by default. The
+  collection, every feature and the persisted Export metadata carry a stable
+  result-trust classification; AI confidence is never represented as truth.
+- Added an explicit `operational` export intent that fails closed unless a
+  persisted authoritative building-reference QA proves exact tile coverage,
+  compatible time, supported geometry, zero warnings, zero false positives
+  and zero false negatives.
+- Updated the Downloads copy so users no longer see an unqualified
+  "Gebouwdetecties" export label.
+
+### Verified
+
+- Focused export and detection-QA suite: 24 passed.
+- Scoped Ruff check: passed.
+- Frontend TypeScript check: passed.
+
+### Remaining limitations
+
+- The active model remains review-required and Kempen-scoped. The export gate
+  prevents overstated results; it does not replace the missing independent,
+  human-reviewed national accuracy corpus or improve model weights by itself.

docs/TODO.md

@@ -1120,7 +1120,9 @@ This file now starts with the current implementation status. Older preparation/b
   with a model/checksum-bound geometry manifest; equivalent legal-scope checks
   still require the same review.
 - [ ] P2-04: make derived persistence transactional, require complete
-  RunManifest hashes and expose every fallback/persistence failure.
+  RunManifest hashes and expose every fallback/persistence failure. Detection
+  exports now carry a machine-readable trust contract and fail closed for
+  operational use without zero-error authoritative QA.
 - [ ] P2-05: remove every protected-test feedback path, introduce a test vault
   and make the sampler reject protected IDs, paths and assessment fields.
 - [ ] P2-06: complete representative human V56 review; add independent AOIs,