Expand YOLO training AOIs safely

backend/tests/test_sprint130_operator_yolo_tile_dataset.py

@@ -6,6 +6,7 @@ import subprocess
 import sys
 
 import numpy as np
+import pytest
 
 
 ROOT = Path(__file__).resolve().parents[2]
@@ -75,6 +76,30 @@ def test_operator_yolo_tile_dataset_export_help_does_not_require_gis_dependencie
     assert "--blank-range-threshold" in result.stdout
 
 
+def test_default_validation_split_is_explicit_and_rejects_holdout_leakage() -> None:
+    module = load_tile_exporter()
+    samples = [
+        {"sample_slug": "geel", "recommended_split": "train"},
+        {"sample_slug": "turnhout", "recommended_split": "val"},
+        {"sample_slug": "retie", "recommended_split": "val"},
+        {"sample_slug": "westerlo", "recommended_split": "val"},
+        {"sample_slug": "arendonk_heide", "recommended_split": "val"},
+    ]
+
+    assert module.DEFAULT_VALIDATION_SAMPLE_SLUGS == frozenset(
+        {"turnhout", "retie", "westerlo", "arendonk_heide"}
+    )
+    assert module.validate_validation_split(
+        samples,
+        set(module.DEFAULT_VALIDATION_SAMPLE_SLUGS),
+    ) == set(module.DEFAULT_VALIDATION_SAMPLE_SLUGS)
+
+    with pytest.raises(SystemExit, match="recommended validation holdouts"):
+        module.validate_validation_split(samples, {"turnhout"})
+    with pytest.raises(SystemExit, match="unknown samples"):
+        module.validate_validation_split(samples, {"turnhout", "missing"})
+
+
 def test_iter_tile_windows_covers_edges_without_duplicates() -> None:
     module = load_tile_exporter()
 

backend/tests/test_sprint131_operator_sample_expansion.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import importlib.util
+import math
 from pathlib import Path
 import sys
 
@@ -43,6 +44,40 @@ def test_operator_sample_registry_includes_kempen_reference_and_background_candi
     assert all(module.SAMPLES[slug].sample_role == "background_candidate" for slug in expected_background_slugs)
 
 
+def test_operator_training_expansion_preserves_geographically_separate_holdouts() -> None:
+    module = load_sample_preparer()
+
+    expected_expansion = {"olen_center", "lille_center", "oud_turnhout_center", "kasterlee_center"}
+    expected_holdouts = {"turnhout", "retie", "westerlo", "arendonk_heide"}
+
+    assert module.TRAINING_EXPANSION_SAMPLE_SLUGS == frozenset(expected_expansion)
+    assert module.DEFAULT_VALIDATION_SAMPLE_SLUGS == frozenset(expected_holdouts)
+    assert all(module.SAMPLES[slug].sample_role == "reference" for slug in expected_expansion)
+    assert all(not module.SAMPLES[slug].allow_empty_reference for slug in expected_expansion)
+    assert all(module.recommended_split_for_sample(module.SAMPLES[slug]) == "train" for slug in expected_expansion)
+    assert all(module.recommended_split_for_sample(module.SAMPLES[slug]) == "val" for slug in expected_holdouts)
+
+    def distance_m(left, right) -> float:
+        radius_m = 6_371_008.8
+        left_lat = math.radians(left.center_lat)
+        right_lat = math.radians(right.center_lat)
+        delta_lat = right_lat - left_lat
+        delta_lon = math.radians(right.center_lon - left.center_lon)
+        haversine = (
+            math.sin(delta_lat / 2) ** 2
+            + math.cos(left_lat) * math.cos(right_lat) * math.sin(delta_lon / 2) ** 2
+        )
+        return 2 * radius_m * math.asin(math.sqrt(haversine))
+
+    reference_holdouts = expected_holdouts - {"arendonk_heide"}
+    for expansion_slug in expected_expansion:
+        expansion = module.SAMPLES[expansion_slug]
+        assert min(
+            distance_m(expansion, module.SAMPLES[holdout_slug])
+            for holdout_slug in reference_holdouts
+        ) >= 2_000
+
+
 def test_operator_background_candidates_are_unique_enough_for_hard_negative_training() -> None:
     module = load_sample_preparer()
 

backend/tests/test_sprint156_background_corpus_classification.py

@@ -90,6 +90,7 @@ def test_prepare_sample_manifest_records_background_category_from_cached_referen
     prepared = module.prepare_sample(sample, tmp_path, force=False)
 
     assert prepared["background_category"] == "sparse_building_context"
+    assert prepared["recommended_split"] == "train"
     assert prepared["reference_feature_count"] == 1
 
 

backend/tests/test_sprint170_detection_false_negative_audit.py

@@ -201,3 +201,62 @@ def test_false_negative_audit_finds_persistent_reference_misses(tmp_path: Path)
     assert active["false_negative_area_m2"]["median"] > 0
     assert report["recommendations"]
     assert (output_dir / "detection_false_negative_audit.md").is_file()
+
+
+def test_false_negative_audit_rejects_mismatched_reference_populations(tmp_path: Path) -> None:
+    script = ROOT / "scripts" / "audit_detection_false_negative_evidence.py"
+    portfolio_args = []
+    for label, source_ids in (("active", ("one", "two")), ("candidate", ("one",))):
+        portfolio_dir = tmp_path / label
+        evidence_dir = portfolio_dir / "samples" / "geel" / "evidence"
+        evidence_dir.mkdir(parents=True)
+        evidence_path = evidence_dir / "calibration_evidence.geojson"
+        evidence_path.write_text(
+            json.dumps(
+                {
+                    "type": "FeatureCollection",
+                    "features": [
+                        _evidence_feature(
+                            "false_negative",
+                            source_id,
+                            _polygon(5.0 + index * 0.001, 51.2, 0.0001),
+                        )
+                        for index, source_id in enumerate(source_ids)
+                    ],
+                }
+            ),
+            encoding="utf-8",
+        )
+        portfolio_path = portfolio_dir / "calibration_evidence_portfolio.json"
+        portfolio_path.write_text(
+            json.dumps(
+                {
+                    "model_asset_id": f"model-{label}",
+                    "samples": [
+                        {
+                            "sample_slug": "geel",
+                            "evidence_geojson_path": str(evidence_path),
+                        }
+                    ],
+                }
+            ),
+            encoding="utf-8",
+        )
+        portfolio_args.extend(("--portfolio", f"{label}={portfolio_path}"))
+
+    result = subprocess.run(
+        [
+            sys.executable,
+            str(script),
+            *portfolio_args,
+            "--output-dir",
+            str(tmp_path / "audit"),
+        ],
+        cwd=ROOT,
+        check=False,
+        capture_output=True,
+        text=True,
+    )
+
+    assert result.returncode != 0
+    assert "different reference populations" in result.stderr