Record live Mol operational evidence

docs/CODEX_EXECUTION_LOG.md

@@ -7287,3 +7287,17 @@ Open:
 - `git diff --check`: clean apart from the existing Windows line-ending notice for the all-in-one Dockerfile.
 - `bash scripts/run_readiness_check.sh`: passed with `485` backend tests, contract audit, single Alembic head, frontend typecheck and production build.
 - Frontend bundle sizes are unchanged: app `217.69 kB`, React vendor `140.74 kB` and MapLibre `801.82 kB` before gzip.
+
+## Live Tower completion
+
+- Deployed through Gitea commit `720ad71`; the all-in-one browser/API smoke, PostGIS `3.6`, required runtime schema and Alembic head `202606120900` passed.
+- Prepared real 1024 px orthophoto/GRB samples for Mol center, Achterbos, Gompel, Donk, Postel and Postel-bos. Positive reference counts are `1,993`, `1,388`, `1,070`, `1,234` and `137`; the background control contains exactly `0` GRB buildings.
+- Fixed two manifest-path type errors found by the first live multi-sample aggregation and moved all-in-one evidence defaults from the replaceable container layer to `/app/storage/operator-evidence`.
+- Completed the persistent four-zone configured-YOLO benchmark at confidence `0.15`, tile `512`, overlap `64` and QA IoU `0.25`: mean precision `0.6286`, mean recall `0.5370`, mean F1 `0.5768`, `2,239` matches, `1,215` false positives and `1,590` false negatives.
+- Per-zone F1 is Achterbos `0.6433`, Gompel `0.6333`, Donk `0.5823` and Postel `0.4483`. Postel is the explicit weakest-zone review priority; no automatic promotion or retraining decision was made.
+- The real empty Postel-bos control produced `0` detections and therefore `0.0` false-positive pressure without fabricated QA metrics.
+- Verified through the browser that the complete Postel project wins initial selection over the newer raster-only background project, with one AOI, one ready raster, one ready GRB reference, one detection run, one QualityCheck and one export.
+- MapLibre rendered the OSM road basemap, AOI and `95` persisted detections without browser warnings or horizontal overflow at `1280x720` and `2560x1080`; the ultrawide canvas measured `1772x830` CSS pixels.
+- Ran the interactive PostGIS AOI query against `vector_features`: `126` GRB buildings returned with no truncation. The guided GIS smoke persisted a derived dataset and GeoJSON export, then produced candidate/reference QA F1 `0.9582` with `126` matches, `0` false positives and `11` false negatives; `263` persisted evidence features rendered back on the map.
+- Remaining evidence caveat: the dataset QA result reports weak CRS-assumption warnings, so those geometry metrics remain explicitly approximate until CRS provenance handling is reviewed.
+- Host observation outside GeoIntel: Unraid recovered after reboot and serves the app, but still reports one disabled/invalid array device. Storage administration should resolve that independently of application development.

docs/TODO.md

@@ -20,6 +20,9 @@ This file now starts with the current implementation status. Older preparation/b
 
 - [x] Make Mol the primary workbench, AOI and operator-sample context while preserving broader Kempen coverage.
 - [x] Add a Mol multi-zone operational pack with independent positive holdouts, background control and persisted map-ready AOIs.
+- [x] Execute the Mol pack on live Tower/PostGIS with real orthophotos, GRB references, configured YOLO, persisted QA/QC and a zero-building background control.
+- [x] Persist combined Mol operator evidence under the Unraid storage mount so reports survive all-in-one container replacement.
+- [ ] Visually review Mol Postel and Donk false-positive/false-negative evidence, classify the dominant error modes and only then decide whether another model-training pass is justified.
 - [x] Backend FastAPI foundation, health endpoint and service structure.
 - [x] React/TypeScript frontend foundation and MapLibre workbench.
 - [x] Map layer visibility, opacity and feature property inspection.