Build governed Belgian training corpus pipeline

scripts/export_operator_yolo_tile_dataset.py

@@ -484,6 +484,8 @@ def export_sample_tiles(
 ) -> list[dict[str, Any]]:
     sample_slug = str(sample["sample_slug"])
     sample_role = str(sample.get("sample_role") or "reference")
+    sample_reference_source = str(sample.get("reference_source") or reference_source).strip().lower()
+    sample_reference_layer = str(sample.get("reference_layer") or reference_layer).strip().lower()
     background_category = background_category_for_sample(sample)
     recommended_split = str(sample.get("recommended_split") or "")
     split = "val" if sample_slug.lower() in val_slugs else "train"
@@ -500,8 +502,8 @@ def export_sample_tiles(
             reference_path,
             dataset,
             min_label_px=min_label_px,
-            reference_source=reference_source,
-            reference_layer=reference_layer,
+            reference_source=sample_reference_source,
+            reference_layer=sample_reference_layer,
         )
         for tile_index, tile_window in enumerate(iter_tile_windows(dataset.width, dataset.height, tile_size, stride)):
             labels = labels_for_tile(