Merge nucleic/sleek-ember-seal-uady into dev

prepare_data.py

@@ -28,6 +28,7 @@ from purpose_data import (
     write_json,
     write_jsonl,
 )
+from review_contract import apply_completed_human_review
 
 
 SCRIPT_DIR = Path(__file__).resolve().parent
@@ -121,6 +122,8 @@ def _build_manifest(
     seed: int,
     near_duplicate_threshold: float,
     curation_review_path: Path | None,
+    human_review_path: Path | None,
+    human_review_summary: dict[str, Any] | None,
 ) -> dict[str, Any]:
     manifest = {
         "schemaVersion": 1,
@@ -181,6 +184,12 @@ def _build_manifest(
     if curation_review_path is not None:
         manifest["curation"]["reviewPath"] = _relative(curation_review_path)
         manifest["curation"]["reviewSha256"] = file_sha256(curation_review_path)
+    if human_review_path is not None:
+        manifest["curation"]["humanReview"] = {
+            "path": _relative(human_review_path),
+            "sha256": file_sha256(human_review_path),
+            "summary": human_review_summary,
+        }
     return manifest
 
 
@@ -195,6 +204,7 @@ def prepare(
     seed: int,
     near_duplicate_threshold: float,
     curation_review_path: Path | None = None,
+    human_review_path: Path | None = None,
 ) -> dict[str, Any]:
     records = load_sources(sources)
     fixtures = load_classifiable_fixtures(fixtures_path)
@@ -215,8 +225,18 @@ def prepare(
         if curation_review_path is not None
         else CurationResult(records=lexical_curation.records, duplicates=[])
     )
+    human_review_summary = None
+    reviewed_records = reviewed_curation.records
+    if human_review_path is not None:
+        human_review = apply_completed_human_review(
+            human_review_path,
+            reviewed_records,
+            dataset_version=DATASET_VERSION,
+        )
+        reviewed_records = human_review.records
+        human_review_summary = human_review.summary
     curated = CurationResult(
-        records=reviewed_curation.records,
+        records=reviewed_records,
         duplicates=lexical_curation.duplicates + reviewed_curation.duplicates,
     )
     splits = split_records(
@@ -283,6 +303,8 @@ def prepare(
         seed=seed,
         near_duplicate_threshold=near_duplicate_threshold,
         curation_review_path=curation_review_path,
+        human_review_path=human_review_path,
+        human_review_summary=human_review_summary,
     )
     # The frozen path can be overridden in tests or experiments.
     manifest["frozenEval"]["syntheticPath"] = _relative(frozen_test_path)
@@ -328,6 +350,14 @@ def build_parser() -> argparse.ArgumentParser:
         default=DEFAULT_CURATION_REVIEW,
         help="completed semantic duplicate review applied after lexical deduplication",
     )
+    parser.add_argument(
+        "--human-review",
+        type=Path,
+        help=(
+            "completed label-and-difficulty review ledger from review_data.py; "
+            "omitted until human review is complete"
+        ),
+    )
     return parser
 
 
@@ -349,6 +379,9 @@ def main(argv: Sequence[str] | None = None) -> int:
             seed=args.seed,
             near_duplicate_threshold=args.near_duplicate_threshold,
             curation_review_path=args.curation_review.resolve(),
+            human_review_path=(
+                args.human_review.resolve() if args.human_review is not None else None
+            ),
         )
     except (DataError, OSError, UnicodeError, json.JSONDecodeError) as exc:
         print(f"error: {exc}", file=sys.stderr)