Required formal scoresA0 avg@3 · A_last avg@3 · five RSI steps
VerifiedRunningPending
Submission console
One review surface for every formal model–harness cell. Results enter through GitHub submissions to rsibench-data; the sections below bind every published score to its trajectory, GitHub submitter, pull request, and merge commit.
Main experiment leaderboard
Rows are the sixteen planned model–harness configurations in the same order as the main leaderboard. Cells show the latest final score with RSI lift in parentheses; ing marks an active formal run. RSIBench does not average incompatible benchmarks into an overall score.