block checkpoint evaluation on ancestral exposure

scripts/README.md

@@ -5,13 +5,16 @@ weights on one declared, non-protected `val` split using CUDA. It records exact
 model and dataset hashes, standard Ultralytics detection metrics and a separate
 pure-background detection count. The output claim is validation ranking only;
 the script neither reads protected test data nor promotes a model.
-Governed comparisons must pass every candidate's exact tile-summary with
-`--training-summary` and enable `--require-training-sample-independence`. If
-any validation AOI occurs in any supplied train split, the script writes a
+Governed comparisons must pass the exact tile summary for every corpus in the
+complete ancestry of every candidate with `--lineage-summary` and enable
+`--require-lineage-sample-independence`. The check covers train, validation,
+calibration and every other recorded split, because an AOI used for ancestral
+checkpoint or threshold selection is also exposed. Any exposure writes a
 blocked manifest and exits before importing PyTorch, loading a model or using
-the GPU. A background detection count from a training-seen AOI is only a
-regression check and must never be presented as independent background
-evidence.
+the GPU. The older `--training-summary` and
+`--require-training-sample-independence` spellings remain aliases, but now use
+the same safer full-split semantics. A result from any lineage-exposed AOI is
+only a regression check and must never be presented as independent evidence.
 
 `render_operator_yolo_label_qa_contact_sheets.py` paginates complete visual
 reviews with `--tiles-per-sheet` (default `64`). This keeps large corpora