Re-run only what matters. Start runs with failed and changed tests for faster feedback and smarter results.