Released EvalSeal on PyPI: repeated LLM eval runs, flip rates, evaluator fingerprints, and CI gates that refuse to compare runs graded differently.