Definition
Plain language
A statistic measuring how well two rankings of the same things agree.
As stated in the literature
A non-parametric rank-correlation coefficient measuring monotonic association between two variables.
Also called: Spearman
Why it matters: It's robust to outliers and non-linear scales, which makes it the right choice when comparing rankings produced by different scoring conventions.
For example, if movie critics and audiences rank the same ten films in nearly the same order, Spearman correlation is close to 1 even if their numerical scores look very different.
Heard on the show
“And the headline empirical result is that this proxy correlates positively with final task performance — Spearman around plus zero-point-three-nine.”Episode 010 — When Reward Climbs But Reasoning Goes Generic: Diagnosing Template Collapse in Agentic RL