document schema seed

README.md

@@ -26,3 +26,5 @@ Public store of keep/drop rulings from the annotation review app (`human_labelin
 from datasets import load_dataset
 ds = load_dataset("rafmacalaba/data-use-annotations")  # train = rulings.jsonl
 ```
+
+Note: `rulings/_schema.jsonl` is a schema placeholder so the dataset always loads — filter `annotator != "_schema"` in analysis.