jev-regress-bench

Which of an agent's approved answers actually changed after a config edit? A reproducible bench: exact match vs embeddings vs a three-tier stack vs an LLM judge vs one Jev Choice.

unverified — listed by yibie/awesome-jev; repository confirmed via the GitHub API, but GitHub code search did not locate Jev code, so the integration has not been confirmed