Initial public release

backend/tests/test_sprint158_promotion_report_split_background.py

@@ -0,0 +1,174 @@
+from __future__ import annotations
+
+import json
+import subprocess
+from pathlib import Path
+
+
+ROOT = Path(__file__).resolve().parents[2]
+
+
+def write_hard_negative_summary(
+    path: Path,
+    *,
+    category: str,
+    detection_counts: list[int],
+) -> None:
+    items = [
+        {
+            "sample_slug": f"{category}_{index}",
+            "background_category": category,
+            "model_asset_id": "candidate-context-sensitive",
+            "tile_size": 512,
+            "tile_overlap": 64,
+            "threshold": 0.35,
+            "tile_count": 4,
+            "detection_count": detection_count,
+            "false_positive_pressure": detection_count / 4,
+        }
+        for index, detection_count in enumerate(detection_counts, start=1)
+    ]
+    path.write_text(
+        json.dumps(
+            {
+                "generated_at": "2026-07-10T00:00:00+00:00",
+                "background_category_counts": {category: len(items)},
+                "items": items,
+            }
+        ),
+        encoding="utf-8",
+    )
+
+
+def test_promotion_report_uses_split_pure_empty_as_gate_and_sparse_context_as_review(
+    tmp_path: Path,
+) -> None:
+    script_path = ROOT / "scripts" / "build_detection_model_promotion_report.py"
+    positive_path = tmp_path / "positive_portfolio.json"
+    pure_summary_path = tmp_path / "pure_empty_summary.json"
+    sparse_summary_path = tmp_path / "sparse_context_summary.json"
+    split_summary_path = tmp_path / "background_corpus_split_summary.json"
+    output_dir = tmp_path / "promotion-report"
+
+    positive_path.write_text(
+        json.dumps(
+            {
+                "items": [
+                    {
+                        "sample_slug": "geel",
+                        "model_asset_id": "candidate-context-sensitive",
+                        "tile_size": 512,
+                        "tile_overlap": 64,
+                        "threshold": 0.35,
+                        "precision": 0.72,
+                        "recall": 0.5,
+                        "f1_score": 0.59,
+                    },
+                    {
+                        "sample_slug": "mol",
+                        "model_asset_id": "candidate-context-sensitive",
+                        "tile_size": 512,
+                        "tile_overlap": 64,
+                        "threshold": 0.35,
+                        "precision": 0.68,
+                        "recall": 0.48,
+                        "f1_score": 0.56,
+                    },
+                ]
+            }
+        ),
+        encoding="utf-8",
+    )
+    write_hard_negative_summary(
+        pure_summary_path,
+        category="pure_empty_negative",
+        detection_counts=[0, 0],
+    )
+    write_hard_negative_summary(
+        sparse_summary_path,
+        category="sparse_building_context",
+        detection_counts=[4, 7],
+    )
+    split_summary_path.write_text(
+        json.dumps(
+            {
+                "schema_version": 1,
+                "source_summaries": {
+                    "pure_empty_negative": str(pure_summary_path),
+                    "sparse_building_context": str(sparse_summary_path),
+                },
+                "strict_default_gate": {
+                    "category": "pure_empty_negative",
+                    "review_only": False,
+                    "sample_count": 2,
+                    "run_count": 2,
+                    "total_detection_count": 0,
+                    "max_detection_count": 0,
+                    "passes_zero_detection_gate": True,
+                },
+                "context_review": {
+                    "category": "sparse_building_context",
+                    "review_only": True,
+                    "sample_count": 2,
+                    "run_count": 2,
+                    "total_detection_count": 11,
+                    "max_detection_count": 7,
+                },
+            }
+        ),
+        encoding="utf-8",
+    )
+
+    result = subprocess.run(
+        [
+            "python",
+            str(script_path),
+            "--positive-portfolio",
+            str(positive_path),
+            "--background-split-summary",
+            str(split_summary_path),
+            "--output-dir",
+            str(output_dir),
+            "--min-positive-samples",
+            "2",
+            "--min-background-samples",
+            "2",
+            "--min-mean-f1",
+            "0.5",
+            "--max-background-detections-per-sample",
+            "0",
+        ],
+        cwd=ROOT,
+        check=True,
+        text=True,
+        capture_output=True,
+    )
+
+    assert "Detection model promotion report passed" in result.stdout
+    report = json.loads((output_dir / "detection_model_promotion_report.json").read_text(encoding="utf-8"))
+    decision = report["candidate_decisions"][0]
+
+    assert report["hard_negative_summary_paths"] == [str(pure_summary_path)]
+    assert report["background_split_summary_paths"] == [str(split_summary_path)]
+    assert report["background_context_reviews"] == [
+        {
+            "source_split_summary_path": str(split_summary_path),
+            "source_summary_path": str(sparse_summary_path),
+            "category": "sparse_building_context",
+            "review_only": True,
+            "sample_count": 2,
+            "run_count": 2,
+            "total_detection_count": 11,
+            "max_detection_count": 7,
+        }
+    ]
+    assert decision["candidate_key"] == "candidate-context-sensitive|512|64|0.35"
+    assert decision["background_sample_count"] == 2
+    assert decision["max_background_detections"] == 0
+    assert decision["promotion_status"] == "promote_candidate"
+    assert report["recommended_candidate"]["candidate_key"] == decision["candidate_key"]
+
+    markdown = (output_dir / "detection_model_promotion_report.md").read_text(encoding="utf-8")
+    assert "Background split summaries: 1" in markdown
+    assert "Sparse-context review evidence" in markdown
+    assert "not used as a default-promotion gate" in markdown