evaluate models on fresh regional calibration AOIs

tests/test_evaluate_yolo_checkpoint_matrix.py

@@ -6,6 +6,7 @@ from scripts.evaluate_yolo_checkpoint_matrix import (
     dataset_overlap_evidence,
     main,
     model_lineage_independence_evidence,
+    validate_pure_background_prefixes,
     write_blocked_manifest,
 )
 
@@ -196,3 +197,32 @@ def test_cli_blocks_lineage_validation_exposure_before_model_resolution(
     payload = json.loads(output.read_text(encoding="utf-8"))
     assert payload["status"] == "blocked_model_lineage_sample_exposure"
     assert payload["model_loading_attempted"] is False
+
+
+def test_pure_background_prefix_rejects_nonempty_labels(tmp_path: Path) -> None:
+    image_dir = tmp_path / "images" / "val"
+    label_dir = tmp_path / "labels" / "val"
+    image_dir.mkdir(parents=True)
+    label_dir.mkdir(parents=True)
+    image = image_dir / "sparse_bg_0001.png"
+    image.write_bytes(b"image")
+    (label_dir / "sparse_bg_0001.txt").write_text(
+        "0 0.5 0.5 0.1 0.1\n", encoding="utf-8"
+    )
+
+    import pytest
+
+    with pytest.raises(ValueError, match="non-empty labels"):
+        validate_pure_background_prefixes([image], ("sparse_bg",))
+
+
+def test_pure_background_prefix_accepts_empty_labels(tmp_path: Path) -> None:
+    image_dir = tmp_path / "images" / "val"
+    label_dir = tmp_path / "labels" / "val"
+    image_dir.mkdir(parents=True)
+    label_dir.mkdir(parents=True)
+    image = image_dir / "pure_bg_0001.png"
+    image.write_bytes(b"image")
+    (label_dir / "pure_bg_0001.txt").write_text("", encoding="utf-8")
+
+    assert validate_pure_background_prefixes([image], ("pure_bg",)) == [image]