Record paged AOI1024 runtime evidence

CHANGELOG.md

@@ -13,6 +13,8 @@
 - Added `--reference-page-limit` / `OPERATOR_GRB_PAGE_LIMIT` and `--reference-max-features` / `OPERATOR_GRB_MAX_FEATURES` safeguards for dense reference AOIs.
 - Generated reference GeoJSON now records fetched page URLs, page count, truncation state and paging limits for auditability.
 - Added regression coverage for paged GRB responses and the new CLI help options.
+- Redeployed Tower, regenerated AOI1024 operator samples with paged GRB references and re-exported `yolo-building-aoi1024-visible025`; dense reference counts now exceed the old cap where expected, including Geel 2,268, Mol 1,993, Turnhout 3,278, Herentals 2,478, Balen 1,343, Retie 1,734 and Westerlo 1,133 features.
+- The refreshed AOI1024 tile dataset now has 29,170 labels with 0 missing label files and 0 invalid rows; audit status remains `needs_attention` because small-box share is still high.
 - No application provider endpoint, migration, API contract, live GRB product import, model download or active YOLO model changed.
 
 ## Sprint 151 runtime GIS upload and AOI1024 YOLO candidate (2026-07-09)

docs/CODEX_EXECUTION_LOG.md

@@ -5954,15 +5954,39 @@ Open:
 - RED: `python -m pytest backend\tests\test_sprint131_operator_sample_expansion.py::test_fetch_reference_follows_grb_next_links_until_complete -q` failed because only the first GRB page was persisted.
 - GREEN: `python -m pytest backend\tests\test_sprint131_operator_sample_expansion.py::test_fetch_reference_follows_grb_next_links_until_complete -q`
 - `python -m pytest backend\tests\test_sprint127_operator_sample_quality_matrix.py backend\tests\test_sprint131_operator_sample_expansion.py -q`
+- `python -m py_compile scripts\prepare_operator_real_data_samples.py`
+- `bash scripts/run_readiness_check.sh`
+- Local live GRB paging smoke with `page_limit=2` and `max_features=5` fetched 3 source pages, returned 5 features and correctly marked `reference_truncated=true`.
+- Tower deploy from commit `a63d4ea` passed live migration smoke and browser runtime verification on `http://192.168.10.150:1202`.
+- Tower one-sample Geel AOI1024 paging smoke passed with `/opt/geointel/venv/bin/python` and produced 2,268 GRB reference features instead of the old 1,000-feature cap.
+- Tower full AOI1024 operator sample regeneration passed:
+  - Geel: 2,268 features, 3 GRB pages, not truncated.
+  - Mol: 1,993 features, 2 GRB pages, not truncated.
+  - Turnhout: 3,278 features, 4 GRB pages, not truncated.
+  - Herentals: 2,478 features, 3 GRB pages, not truncated.
+  - Balen: 1,343 features, 2 GRB pages, not truncated.
+  - Retie: 1,734 features, 2 GRB pages, not truncated.
+  - Westerlo: 1,133 features, 2 GRB pages, not truncated.
+- Re-exported `/app/storage/operator-data/yolo-building-aoi1024-visible025` from the regenerated paged references:
+  - 144 tiles, 117 positive tiles, 27 negative tiles, 29,170 labels.
+  - `min_label_visible_ratio=0.25`, `negative_keep_ratio=1.0`.
+- Re-ran the Tower dataset audit:
+  - output: `/app/storage/operator-data/yolo-building-aoi1024-visible025-audit/operator_yolo_dataset_quality_audit.json`
+  - status remains `needs_attention`
+  - missing label files: 0
+  - invalid labels: 0
+  - median normalized box area: `0.000762939453125`
+  - small-box share: `0.34744600617072335`
+  - repeated background negative share: 0.0
 
 ## Known limitations
 
-- Existing Tower AOI1024 artifacts were generated before this fix. Regenerate the operator samples with `--force` before retraining or re-auditing dense AOI labels.
+- The Tower AOI1024 operator samples and tile dataset have now been regenerated from paged GRB references, but the tile audit still reports `needs_attention` because small clipped boxes remain common.
 - This does not make the Sprint 7B GRB provider a live product importer; it only fixes explicit operator sample preparation.
 
 ## Next recommended pass
 
-- Regenerate `/app/storage/operator-data/operator-samples-1024` on Tower with the paged script, re-export the AOI1024 YOLO tile dataset, rerun the dataset audit and only then consider another inactive training candidate.
+- Before another training run, improve label quality: either raise `--min-label-visible-ratio`, increase tile size, reduce dense clipped-edge tiles or add cleaner AOIs. Do not activate or retrain from this dataset as-is without acknowledging the small-box warning.
 
 # Sprint 151 - Runtime GIS upload and AOI1024 YOLO candidate
 

docs/TODO.md

@@ -473,5 +473,6 @@ This file now starts with the current implementation status. Older preparation/b
 - [x] Fix promotion-report parsing for `multi_sample_quality_summary.json` inputs.
 - [x] Generate AOI1024 promotion report and keep recommended candidate as `none`.
 - [x] Add GRB paging before trusting dense reference exports as full ground truth.
-- [ ] Regenerate Tower AOI1024 operator samples with paged GRB references, then re-export and audit labels before any new training attempt.
+- [x] Regenerate Tower AOI1024 operator samples with paged GRB references, then re-export and audit labels before any new training attempt.
+- [ ] Improve AOI1024 label quality before retraining: current paged export has 29,170 labels and no invalid rows, but audit remains `needs_attention` because small-box share is `0.34744600617072335`.
 - [ ] Keep every local YOLO candidate inactive until positive-AOI and hard-negative promotion reports recommend default activation.

scripts/README.md

@@ -396,12 +396,15 @@ Current Tower audit status:
 - `yolo-building-tile-uniquehardneg160`: clean expanded-background baseline;
   576 tiles, 346 positive, 230 negative, 11,757 labels, 0 invalid labels and
   0 repeated background negatives in the first Tower audit.
-- `yolo-building-aoi1024-visible025`: larger AOI candidate baseline; 144
-  tiles, 117 positive tiles, 27 negative tiles, 15,079 labels and
+- `yolo-building-aoi1024-visible025`: larger AOI candidate baseline regenerated
+  after paged GRB references; 144 tiles, 117 positive tiles, 27 negative tiles,
+  29,170 labels, 0 missing label files, 0 invalid labels and
   `min_label_visible_ratio=0.25`. Its audit remains `needs_attention` because
-  the median normalized box area is still small. The trained
-  `geointel-building-yolov8s-aoi1024visible025e50-pt` asset is inactive until
-  persisted QA/QC and hard-negative promotion gates pass.
+  the median normalized box area is still small (`0.000762939453125`) and the
+  small-box share is high (`0.34744600617072335`). The previously trained
+  `geointel-building-yolov8s-aoi1024visible025e50-pt` asset is inactive; do not
+  retrain or activate from this dataset without addressing label quality or
+  explicitly accepting the small-box warning.
 
 After rebuilding the all-in-one image, the operator scripts are available inside
 the container at `/app/scripts/...`. Before rebuilding, use the host checkout or