Require tile quality evidence for training

scripts/run_belgium_building_training_loop.py

@@ -29,7 +29,10 @@ def write_json(path: Path, value: dict[str, Any]) -> None:
     temporary.replace(path)
 
 
-def dataset_audit_failures(audit: dict[str, Any]) -> list[str]:
+def dataset_audit_failures(
+    audit: dict[str, Any],
+    train_quality_audit: dict[str, Any],
+) -> list[str]:
     """Return automated corpus blockers while leaving final human review deferred."""
     failures = [str(item) for item in audit.get("failures") or []]
     status = audit.get("status")
@@ -39,7 +42,13 @@ def dataset_audit_failures(audit: dict[str, Any]) -> list[str]:
         failures.append("corpus manifest is not immutable")
     if audit.get("spatial_leakage_status") != "ok":
         failures.append("spatial leakage audit is not ok")
-    if int(audit.get("low_variance_positive_tile_count") or 0) != 0:
+    if train_quality_audit.get("status") != "ok":
+        failures.append("train tile quality audit is not ok")
+    if int(train_quality_audit.get("label_stats", {}).get("invalid_label_count", -1)) != 0:
+        failures.append("train tile quality audit contains invalid labels")
+    if int(train_quality_audit.get("label_stats", {}).get("missing_label_file_count", -1)) != 0:
+        failures.append("train tile quality audit contains missing label files")
+    if int(train_quality_audit.get("low_variance_positive_tile_count", -1)) != 0:
         failures.append("dataset contains blank/low-variance positive tiles")
     return failures
 
@@ -188,6 +197,7 @@ def main() -> int:
     parser.add_argument("--train-yaml", type=Path, required=True)
     parser.add_argument("--train-summary", type=Path, required=True)
     parser.add_argument("--dataset-audit", type=Path, required=True)
+    parser.add_argument("--train-quality-audit", type=Path, required=True)
     parser.add_argument("--calibration-summary", type=Path, required=True)
     parser.add_argument("--test-summary", type=Path, required=True)
     parser.add_argument("--background-summary", type=Path, required=True)
@@ -230,7 +240,8 @@ def main() -> int:
     if args.iterations < 1:
         raise SystemExit("--iterations must be positive")
     dataset_audit = json.loads(args.dataset_audit.read_text(encoding="utf-8"))
-    audit_failures = dataset_audit_failures(dataset_audit)
+    train_quality_audit = json.loads(args.train_quality_audit.read_text(encoding="utf-8"))
+    audit_failures = dataset_audit_failures(dataset_audit, train_quality_audit)
     if audit_failures:
         raise SystemExit(f"Dataset audit is not eligible for training: {audit_failures}")
 
@@ -242,6 +253,9 @@ def main() -> int:
         "initial_model": str(args.initial_model),
         "train_yaml": str(args.train_yaml),
         "dataset_audit": str(args.dataset_audit),
+        "dataset_audit_sha256": sha256(args.dataset_audit),
+        "train_quality_audit": str(args.train_quality_audit),
+        "train_quality_audit_sha256": sha256(args.train_quality_audit),
         "corpus_manifest": str(args.corpus_manifest),
         "iterations": [],
     }

scripts/tower-v37-completion-command.json

@@ -13,6 +13,8 @@
   "/app/storage/operator-data/building-be-v30-rotated-holdouts-r1/train/yolo_tile_dataset_summary.json",
   "--dataset-audit",
   "/app/storage/operator-data/building-be-v30-rotated-holdouts-r1/audit/belgium-building-corpus-audit.json",
+  "--train-quality-audit",
+  "/app/storage/operator-data/building-be-v30-rotated-holdouts-r1/train/audit/operator_yolo_dataset_quality_audit.json",
   "--calibration-summary",
   "/app/storage/operator-data/building-be-v30-rotated-holdouts-r1/calibration/yolo_tile_dataset_summary.json",
   "--test-summary",