fix: bound detection QA to inference coverage

docs/AI_PIPELINES.md

@@ -548,9 +548,14 @@ Sprint 8C makes persisted detections reviewable:
 - QA results reuse `quality_checks` and `metrics`; no parallel QA persistence system is introduced.
 - Configured-YOLO QA derives its evaluation extent from the persisted tile
   manifest. Tile bounds are transformed from their explicit source CRS to
-  EPSG:4326, unioned, and used to clip candidate/reference populations before
-  canonical footprint-IoU matching. Reference features wholly outside the
-  imagery presented to the model no longer count as false negatives.
+  EPSG:4326 and unioned. The union is first applied as a GiST-backed PostGIS
+  spatial predicate, then used to clip the bounded candidate/reference
+  populations before canonical footprint-IoU matching. Complete source counts
+  remain in QA evidence, but regional geometries outside inference coverage are
+  not materialized in application memory and do not count as false negatives.
+- Canonical one-to-one IoU matching uses an in-memory spatial index only to
+  discard geometries whose envelopes cannot intersect. It does not change the
+  configured IoU threshold, greedy match ownership or persisted metrics.
 - A separate reference-envelope IoU pass is persisted as
   `box_to_footprint_diagnostics`. It quantifies possible matching artifacts from
   comparing rectangular detections with irregular building footprints, but is

docs/API_CONTRACTS.md

@@ -893,7 +893,11 @@ Response persists a `quality_check` and `metrics` rows through the existing QA/Q
 Configured-YOLO QA automatically reads `tile_manifest_path` from the persisted
 `AnalysisRun.parameters_json`. Candidate and reference geometries are clipped
 to the union of the manifest's tile bounds after explicit CRS transformation to
-EPSG:4326. The response additionally returns:
+EPSG:4326. Before reference geometries are materialized, the service applies
+that coverage with an indexed PostGIS `ST_Intersects` predicate. The full
+dataset count is retained separately so raw/evaluated/excluded counts remain
+auditable without transferring a regional reference dataset to Python. The
+response additionally returns:
 
 - `candidate_feature_count_raw` and `reference_feature_count_raw`;
 - `coverage`, including raw/evaluated/excluded/boundary-clipped population

docs/CODEX_EXECUTION_LOG.md

@@ -8052,3 +8052,19 @@ Live operational proof:
 
 Next:
 - Reuse the bounded regional operator pattern for current roads, water and parcels, then add official population and land-use time series without changing Mol semantics or introducing interactive external fetching.
+
+## Sprint 195 - Live guided detection and bounded regional QA (2026-07-14)
+
+Implemented:
+- Added one guided Detection Lab action that uploads a georeferenced raster through `DatasetService`, creates or reuses the canonical tile manifest, runs preflight, invokes the configured local YOLO adapter and opens persisted detection geometry on MapLibre.
+- Live Tower validation used the official Mol orthophoto sample, produced nine 512 px tiles and persisted 1,953 detections in analysis run `3912e179-3d1d-4080-8093-f883bbe95d9c`.
+- A real QA attempt against the 466,078-feature regional GRB building dataset revealed that coverage clipping happened after every reference row had already been materialized.
+- Moved configured-YOLO reference bounding into the existing GiST-indexed PostGIS query with the persisted manifest coverage. Full source count, evaluated count and excluded count remain explicit in the persisted evidence.
+- Added a Shapely STRtree candidate index around the unchanged exact IoU matcher and clarified the simple map legend/source whenever AI detections are active.
+
+Validation before final deployment:
+- Focused detection, coverage, QA and guided-workflow tests passed.
+- Frontend TypeScript typecheck passed after the source/legend correction.
+
+Next:
+- Deploy the bounded QA correction, rerun the live persisted detection-versus-GRB comparison, verify persisted metrics and inspect the result in MapLibre before declaring the guided operational flow complete.