Restrict holdout overlap check to holdout vs dev/full only

dev and full are allowed to share files — only holdout must be strictly
isolated. When creating dev or full, only compare against holdout.manifest;
when creating holdout, compare against all other manifests in the dir.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

scripts/bump_dataset_version.py

@@ -263,7 +263,13 @@ def check_holdout_overlap(
 
     new_set = set(resolved_new_files)
     overlaps: list[tuple[str, Path]] = []
-    for existing in sorted(manifest_dir.glob("*.manifest")):
+    # When creating holdout, check against all other manifests (dev, full, …).
+    # When creating dev/full, only check against holdout — dev vs full overlap is allowed.
+    if output_resolved.name == "holdout.manifest":
+        candidates = sorted(manifest_dir.glob("*.manifest"))
+    else:
+        candidates = [holdout_path]
+    for existing in candidates:
         if existing.resolve() == output_resolved:
             continue
         try:

tests/test_bump_dataset_version.py

@@ -293,6 +293,21 @@ def test_overlap_detected_when_holdout_already_exists(tmp_path):
     assert overlaps[0][0] == "holdout.manifest"
 
 
+def test_dev_full_overlap_not_checked(tmp_path):
+    pool_dir = tmp_path / "pools" / "pbwo4"
+    pool_dir.mkdir(parents=True)
+    pq = tmp_path / "a.parquet"
+    pq.touch()
+
+    # dev.manifest lists the same file — should not trigger a warning for full
+    dev = pool_dir / "dev.manifest"
+    dev.write_text(os.path.relpath(pq.resolve(), start=pool_dir) + "\n")
+
+    output = pool_dir / "full.manifest"
+    overlaps = check_holdout_overlap(output, [pq.resolve()])
+    assert overlaps == []
+
+
 def test_no_overlap_when_files_are_disjoint(tmp_path):
     pool_dir = tmp_path / "pools" / "pbwo4"
     pool_dir.mkdir(parents=True)