block checkpoint evaluation on ancestral exposure

tests/test_evaluate_yolo_checkpoint_matrix.py

@@ -1,9 +1,11 @@
 import json
+import sys
 from pathlib import Path
 
 from scripts.evaluate_yolo_checkpoint_matrix import (
     dataset_overlap_evidence,
-    training_sample_independence_evidence,
+    main,
+    model_lineage_independence_evidence,
     write_blocked_manifest,
 )
 
@@ -67,7 +69,7 @@ def _write_summary(path: Path, rows: list[tuple[str, str]]) -> None:
     )
 
 
-def test_training_sample_independence_detects_overlap_for_each_corpus(
+def test_model_lineage_independence_detects_exposure_in_every_split(
     tmp_path: Path,
 ) -> None:
     evaluation = tmp_path / "evaluation"
@@ -80,24 +82,25 @@ def test_training_sample_independence_detects_overlap_for_each_corpus(
     )
     active = tmp_path / "active.json"
     challenger = tmp_path / "challenger.json"
-    _write_summary(active, [("postel_bos", "train"), ("mol", "train")])
+    _write_summary(active, [("postel_bos", "train"), ("turnhout", "val")])
     _write_summary(
         challenger, [("postel_bos", "train"), ("dessel", "train")]
     )
 
-    evidence = training_sample_independence_evidence(
+    evidence = model_lineage_independence_evidence(
         dataset_yaml, [active, challenger]
     )
 
     assert evidence["status"] == "overlap"
-    assert evidence["overlapping_evaluation_samples"] == ["postel_bos"]
-    assert evidence["independent_for_all_supplied_training_corpora"] is False
+    assert evidence["overlapping_evaluation_samples"] == ["postel_bos", "turnhout"]
+    assert evidence["independent_for_all_supplied_lineage_corpora"] is False
     assert [
-        row["overlapping_evaluation_samples"] for row in evidence["training_corpora"]
-    ] == [["postel_bos"], ["postel_bos"]]
+        row["overlapping_evaluation_samples"] for row in evidence["lineage_corpora"]
+    ] == [["postel_bos", "turnhout"], ["postel_bos"]]
+    assert evidence["lineage_corpora"][0]["exposure_roles"]["turnhout"] == ["val"]
 
 
-def test_training_sample_independence_accepts_disjoint_samples(tmp_path: Path) -> None:
+def test_model_lineage_independence_accepts_disjoint_samples(tmp_path: Path) -> None:
     evaluation = tmp_path / "evaluation"
     evaluation.mkdir()
     dataset_yaml = evaluation / "dataset.yaml"
@@ -108,14 +111,14 @@ def test_training_sample_independence_accepts_disjoint_samples(tmp_path: Path) -
     training = tmp_path / "training.json"
     _write_summary(training, [("mol", "train")])
 
-    evidence = training_sample_independence_evidence(dataset_yaml, [training])
+    evidence = model_lineage_independence_evidence(dataset_yaml, [training])
 
     assert evidence["status"] == "independent"
     assert evidence["overlapping_evaluation_samples"] == []
-    assert evidence["independent_for_all_supplied_training_corpora"] is True
+    assert evidence["independent_for_all_supplied_lineage_corpora"] is True
 
 
-def test_training_sample_independence_is_unavailable_without_training_corpus(
+def test_model_lineage_independence_is_unavailable_without_ancestral_corpus(
     tmp_path: Path,
 ) -> None:
     evaluation = tmp_path / "evaluation"
@@ -126,11 +129,11 @@ def test_training_sample_independence_is_unavailable_without_training_corpus(
         evaluation / "yolo_tile_dataset_summary.json", [("turnhout", "val")]
     )
 
-    evidence = training_sample_independence_evidence(dataset_yaml, [])
+    evidence = model_lineage_independence_evidence(dataset_yaml, [])
 
     assert evidence["status"] == "unavailable"
     assert evidence["overlapping_evaluation_samples"] == []
-    assert evidence["independent_for_all_supplied_training_corpora"] is False
+    assert evidence["independent_for_all_supplied_lineage_corpora"] is False
 
 
 def test_blocked_manifest_records_that_models_and_gpu_were_not_used(
@@ -145,11 +148,51 @@ def test_blocked_manifest_records_that_models_and_gpu_were_not_used(
         dataset_yaml,
         {
             "status": "overlap",
-            "independent_for_all_supplied_training_corpora": False,
+            "independent_for_all_supplied_lineage_corpora": False,
         },
     )
 
     payload = json.loads(output.read_text(encoding="utf-8"))
-    assert payload["status"] == "blocked_training_sample_overlap"
+    assert payload["status"] == "blocked_model_lineage_sample_exposure"
     assert payload["model_loading_attempted"] is False
     assert payload["gpu_inference_attempted"] is False
+
+
+def test_cli_blocks_lineage_validation_exposure_before_model_resolution(
+    tmp_path: Path, monkeypatch
+) -> None:
+    evaluation = tmp_path / "evaluation"
+    images = evaluation / "images" / "val"
+    images.mkdir(parents=True)
+    (images / "turnhout_r0_c0.png").write_bytes(b"not opened before gate")
+    dataset_yaml = evaluation / "dataset.yaml"
+    dataset_yaml.write_text("path: .\nval: images/val\n", encoding="utf-8")
+    _write_summary(
+        evaluation / "yolo_tile_dataset_summary.json", [("turnhout", "val")]
+    )
+    ancestor = tmp_path / "ancestor.json"
+    _write_summary(ancestor, [("turnhout", "val"), ("mol", "train")])
+    output = tmp_path / "blocked.json"
+    monkeypatch.setattr(
+        sys,
+        "argv",
+        [
+            "evaluate_yolo_checkpoint_matrix.py",
+            "--dataset-yaml",
+            str(dataset_yaml),
+            "--model",
+            str(tmp_path / "model-is-never-resolved.pt"),
+            "--output",
+            str(output),
+            "--background-prefix",
+            "background",
+            "--lineage-summary",
+            str(ancestor),
+            "--require-lineage-sample-independence",
+        ],
+    )
+
+    assert main() == 3
+    payload = json.loads(output.read_text(encoding="utf-8"))
+    assert payload["status"] == "blocked_model_lineage_sample_exposure"
+    assert payload["model_loading_attempted"] is False