No Free Lunch
A theorem stating that, averaged over every possible problem, no learning algorithm outperforms another — an edge on real problems comes from matching their structure.
Averaged uniformly over every possible target function, every learning algorithm's expected performance on unseen data is identical, including against one that predicts randomly. Real problems are never uniform draws over all possible functions — they have structure, and every practical algorithm bakes in an inductive bias that performs well exactly when it matches that structure.
A benchmark win reflects a good match between an algorithm's assumptions and that benchmark's data, not universal superiority across every dataset.