How well a benchmark measures the skill it claims to test.
A kid can ace a driving game at the arcade. That does not mean you hand them car keys.
Benchmark validity asks if a score shows real skill. It shapes leaderboards and model choices.
Benchmark contamination
Benchmark contamination can make scores look too high and harm validity.
Benchmark Saturation
Benchmark Saturation can hide real differences between models.
Leaderboard
Valid benchmarks make a Leaderboard useful.