Blind rank test
Two models from the board, names hidden. Pick the one you think scores higher — then see whether the measurement can tell them apart at all.
93% of the 133 adjacent steps on this board are inside the published error band. Computed from the catalogue on this page load, not written down.
Pair 1 of 8
Scores are LMArena Elo, used under CC BY 4.0 from the official dataset. Nothing you click is recorded.