feat(provenance): govern source snapshots and data inputs

scripts/build_regional_yolo_dataset.py

@@ -6,10 +6,25 @@ from __future__ import annotations
 import argparse
 import hashlib
 import json
+import sys
 from collections import Counter
 from pathlib import Path
 from typing import Any
 
+SCRIPT_DIR = Path(__file__).resolve().parent
+if str(SCRIPT_DIR) not in sys.path:
+    sys.path.insert(0, str(SCRIPT_DIR))
+
+from training_dataset_eligibility import (  # noqa: E402
+    TrainingEligibilityError,
+    assert_frozen_manifest_training_eligible,
+)
+from training_release_manifest import (  # noqa: E402
+    TrainingReleaseError,
+    assert_yolo_summary_bound_to_embedded_training_release,
+    create_training_release_manifest,
+)
+
 
 PROTECTED_SPLITS = {"calibration", "test", "background-test"}
 
@@ -138,11 +153,34 @@ def main() -> int:
     parser.add_argument("--priority-context", action="append", default=[])
     parser.add_argument("--priority-repeat", type=int, default=2)
     parser.add_argument("--negative-repeat", type=int, default=2)
+    parser.add_argument(
+        "--review-audit",
+        type=Path,
+        help="Accepted corpus-audit evidence; mandatory for an operational training release.",
+    )
+    parser.add_argument(
+        "--fixture-mode",
+        action="store_true",
+        help="Accept only an explicitly fixture-only corpus manifest; never use for operational training data.",
+    )
     args = parser.parse_args()
     if args.priority_repeat < 1 or args.negative_repeat < 1:
         parser.error("repeat counts must be positive")
     summary = json.loads(args.summary.read_text(encoding="utf-8"))
     manifest = json.loads(args.corpus_manifest.read_text(encoding="utf-8"))
+    try:
+        assert_frozen_manifest_training_eligible(
+            args.corpus_manifest,
+            fixture_mode=args.fixture_mode,
+            verify_live=True,
+        )
+        assert_yolo_summary_bound_to_embedded_training_release(
+            summary_path=args.summary,
+            corpus_manifest=args.corpus_manifest,
+            fixture_mode=args.fixture_mode,
+        )
+    except (TrainingEligibilityError, TrainingReleaseError) as exc:
+        raise SystemExit(str(exc)) from exc
     train, val, evidence = build(
         summary=summary,
         manifest=manifest,
@@ -161,6 +199,15 @@ def main() -> int:
         f"path: /\ntrain: {train_path}\nval: {val_path}\nnames:\n  0: building\n",
         encoding="utf-8",
     )
+    try:
+        release_paths = create_training_release_manifest(
+            train_yaml=yaml_path,
+            corpus_manifest=args.corpus_manifest,
+            review_audit_path=args.review_audit,
+            fixture_mode=args.fixture_mode,
+        )
+    except TrainingReleaseError as exc:
+        raise SystemExit(str(exc)) from exc
     evidence.update({
         "source_summary": str(args.summary),
         "source_summary_sha256": sha256(args.summary),
@@ -169,6 +216,11 @@ def main() -> int:
         "train_sha256": sha256(train_path),
         "validation_sha256": sha256(val_path),
         "dataset_yaml": str(yaml_path),
+        "training_release_manifest": str(release_paths["release_manifest"]),
+        "training_release_manifest_sha256": sha256(release_paths["release_manifest"]),
+        "training_release_freeze": str(release_paths["release_freeze"]),
+        "training_asset_manifest": str(release_paths["asset_manifest"]),
+        "fixture_mode": bool(args.fixture_mode),
     })
     encoded_evidence = json.dumps(evidence, indent=2)
     (args.output_dir / "regional-dataset-evidence.json").write_text(encoded_evidence, encoding="utf-8")