3.1.1 (2026-07-14) Bug Fixes evals: compute macro/weighted F-score per class to match sklearn semantics (#13740) (c0f6267) evals: count AsyncExecutor timeouts against max_retries (#14361) (90eee08) evals: gate 0/1 positive_label auto-detection on default macro average (#14012) (f96dbd9)