MathArena · BrokenArXiv
MathArena / ETH Zurich / Reasoning · Finding and fixing errors in mathematical proofs
What it measures, and how
Reads raw scores per individual paper and version, without using an Expected performance composite fitted across question banks.
How this evidence is used
Under observation: the re-display boundary for leaderboard scores is not yet clear, so they are not collected into the official score for now.
Limits and data attribution
Under observation: the re-display boundary for leaderboard scores is not yet clear, so they are not collected into the official score for now.
Data licence: Code under MIT; data has a CC BY-NC-SA 4.0 boundary, so question banks and leaderboard scores must be distinguished.
Scores published by MathArena / ETH Zurich. Raw scores and the News consensus score use different scales and cannot be added directly.