Did you experiment with getting an AI to critique incoming PRs, and ignoring ones where it finds clear red flags?