Harden detection result review

backend/tests/test_sprint175_detection_review_hardening.py

@@ -0,0 +1,222 @@
+from __future__ import annotations
+
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+
+ROOT = Path(__file__).resolve().parents[2]
+
+
+def _polygon(x: float, y: float, size: float) -> dict:
+    return {
+        "type": "Polygon",
+        "coordinates": [
+            [
+                [x, y],
+                [x + size, y],
+                [x + size, y + size],
+                [x, y + size],
+                [x, y],
+            ]
+        ],
+    }
+
+
+def _evidence_feature(
+    role: str,
+    feature_id: str,
+    geometry: dict,
+    *,
+    tile_index: int = 1,
+    confidence: float | None = None,
+) -> dict:
+    properties = {
+        "qa_evidence_role": role,
+        "candidate_feature_id": feature_id,
+        "feature_class": "building",
+        "tile_index": tile_index,
+        "analysis_run_id": "run-1",
+        "quality_check_id": "quality-1",
+        "calibration_threshold": 0.15,
+        "calibration_model_asset_id": "model-a",
+    }
+    if confidence is not None:
+        properties["candidate_confidence"] = confidence
+    return {
+        "type": "Feature",
+        "id": f"{role}:{feature_id}",
+        "properties": properties,
+        "geometry": geometry,
+    }
+
+
+def _write_portfolio(
+    tmp_path: Path,
+    features: list[dict],
+    *,
+    declared_false_positives: int,
+) -> Path:
+    evidence_dir = tmp_path / "samples" / "geel" / "evidence"
+    evidence_dir.mkdir(parents=True)
+    evidence_path = evidence_dir / "calibration_evidence.geojson"
+    evidence_path.write_text(
+        json.dumps({"type": "FeatureCollection", "features": features}),
+        encoding="utf-8",
+    )
+    portfolio_path = tmp_path / "calibration_evidence_portfolio.json"
+    portfolio_path.write_text(
+        json.dumps(
+            {
+                "model_asset_id": "model-a",
+                "model_sha256": "abc123",
+                "samples": [
+                    {
+                        "sample_slug": "geel",
+                        "role_counts": {
+                            "false_positive": declared_false_positives,
+                            "match_candidate": 1,
+                        },
+                        "evidence_geojson_path": str(evidence_path),
+                    }
+                ],
+            }
+        ),
+        encoding="utf-8",
+    )
+    return portfolio_path
+
+
+def test_detection_results_table_uses_bounded_local_pagination() -> None:
+    lab = (ROOT / "frontend" / "src" / "components" / "detection" / "DetectionLab.tsx").read_text(
+        encoding="utf-8"
+    )
+    styles = (ROOT / "frontend" / "src" / "styles" / "app.css").read_text(
+        encoding="utf-8"
+    )
+
+    assert "DETECTION_PAGE_SIZE_OPTIONS" in lab
+    assert "visibleDetectionItems" in lab
+    assert "detectionItems.slice" in lab
+    assert "Detection result pagination" in lab
+    assert "Previous detection results page" in lab
+    assert "Next detection results page" in lab
+    assert "pagination-toolbar" in styles
+    assert "detectionItems.map((detection)" not in lab
+
+
+def test_false_positive_audit_builds_reviewable_persisted_evidence(tmp_path: Path) -> None:
+    script = ROOT / "scripts" / "audit_detection_false_positive_evidence.py"
+    readiness = (ROOT / "scripts" / "run_readiness_check.sh").read_text(encoding="utf-8")
+    dockerfile = (ROOT / "deploy" / "unraid" / "Dockerfile.all-in-one").read_text(
+        encoding="utf-8"
+    )
+    assert script.exists()
+    assert "py_compile scripts/audit_detection_false_positive_evidence.py" in readiness
+    assert "COPY scripts/audit_detection_false_positive_evidence.py" in dockerfile
+
+    features = [
+        _evidence_feature(
+            "false_positive",
+            "fp-small",
+            _polygon(5.0, 51.2, 0.0001),
+            tile_index=4,
+            confidence=0.27,
+        ),
+        _evidence_feature(
+            "false_positive",
+            "fp-large",
+            _polygon(5.001, 51.2, 0.0003),
+            tile_index=4,
+            confidence=0.81,
+        ),
+        _evidence_feature(
+            "match_candidate",
+            "matched",
+            _polygon(5.002, 51.2, 0.0002),
+            tile_index=7,
+        ),
+    ]
+    portfolio_path = _write_portfolio(
+        tmp_path / "portfolio",
+        features,
+        declared_false_positives=2,
+    )
+    output_dir = tmp_path / "audit"
+    result = subprocess.run(
+        [
+            sys.executable,
+            str(script),
+            "--portfolio",
+            str(portfolio_path),
+            "--output-dir",
+            str(output_dir),
+        ],
+        cwd=ROOT,
+        check=True,
+        capture_output=True,
+        text=True,
+    )
+
+    report = json.loads(
+        (output_dir / "detection_false_positive_audit.json").read_text(encoding="utf-8")
+    )
+    assert report["model_asset_id"] == "model-a"
+    assert report["model_sha256"] == "abc123"
+    assert report["false_positive_count"] == 2
+    assert report["candidate_count"] == 3
+    assert report["false_positive_rate"] == 2 / 3
+    assert report["confidence_coverage_count"] == 2
+    assert report["confidence"]["median"] == 0.54
+    assert report["source_tile_counts"] == {"geel:4": 2}
+    assert sum(bucket["count"] for bucket in report["area_buckets"].values()) == 2
+    assert report["samples"][0]["false_positive_area_m2"]["median"] > 0
+    assert report["recommendations"]
+
+    geojson = json.loads(
+        (output_dir / "false_positives.geojson").read_text(encoding="utf-8")
+    )
+    assert geojson["type"] == "FeatureCollection"
+    assert len(geojson["features"]) == 2
+    assert all(
+        feature["properties"]["qa_evidence_role"] == "false_positive"
+        for feature in geojson["features"]
+    )
+    assert all(feature["properties"]["sample_slug"] == "geel" for feature in geojson["features"])
+    assert all(feature["properties"]["area_m2"] > 0 for feature in geojson["features"])
+    assert "False-positive audit JSON" in result.stdout
+    assert (output_dir / "detection_false_positive_audit.md").is_file()
+
+
+def test_false_positive_audit_rejects_declared_role_count_drift(tmp_path: Path) -> None:
+    script = ROOT / "scripts" / "audit_detection_false_positive_evidence.py"
+    portfolio_path = _write_portfolio(
+        tmp_path / "portfolio",
+        [
+            _evidence_feature(
+                "false_positive",
+                "fp-one",
+                _polygon(5.0, 51.2, 0.0001),
+            )
+        ],
+        declared_false_positives=2,
+    )
+
+    result = subprocess.run(
+        [
+            sys.executable,
+            str(script),
+            "--portfolio",
+            str(portfolio_path),
+            "--output-dir",
+            str(tmp_path / "audit"),
+        ],
+        cwd=ROOT,
+        check=False,
+        capture_output=True,
+        text=True,
+    )
+
+    assert result.returncode != 0
+    assert "declares 2 false positives but evidence contains 1" in result.stderr