Medical AI Benchmark and Study Review

Field notes

The homepage records 14 benchmarks and studies, assigning each a letter grade alongside a 47–94 score range and a median of 77. Its rubric separates methodological concerns that are often collapsed into a single benchmark score, including model-version reporting, contamination controls, judge calibration, clinical comparators, and deployment framing. The page says reviewer disagreement of 1.5 points or more triggers discussion, while four named failures can impose grade ceilings.