feat(accuracy): build hardened phase 4 evaluation harness

This commit is contained in:
Jens
2026-08-02 02:43:41 +02:00
parent 7b92e29e49
commit bd2fd9780f
13 changed files with 5468 additions and 7 deletions
+4 -3
View File
@@ -35,11 +35,12 @@ def test_regional_dataset_balances_contexts_and_keeps_validation_closed() -> Non
assert evidence["protected_samples_in_training"] == []
def test_regional_dataset_rejects_protected_tiles() -> None:
@pytest.mark.parametrize("protected_split", ["calibration", "test", "background-test", "challenge"])
def test_regional_dataset_rejects_protected_tiles(protected_split: str) -> None:
manifest = {"samples": [
{"sample_slug": "f-cal", "region": "flanders", "split": "calibration", "context": "ribbon"},
{"sample_slug": "f-cal", "region": "flanders", "split": protected_split, "context": "ribbon"},
]}
summary = {"tiles": [{"sample_slug": "f-cal", "split": "calibration", "image_path": "/cal.png"}]}
summary = {"tiles": [{"sample_slug": "f-cal", "split": protected_split, "image_path": "/cal.png"}]}
with pytest.raises(ValueError, match="protected"):
module.build(
summary=summary, manifest=manifest, region="flanders",