Function wilcoxon_worse 

pub fn wilcoxon_worse(
    candidate: &[f64],
    incumbent: &[f64],
    direction: Direction,
    alpha: f64,
    min_pairs: usize,
) -> bool

Whether candidate is significantly worse than incumbent under a paired signed-rank test at level alpha, requiring at least min_pairs effective (non-zero) pairs.

The two slices are paired positionally (element i against element i), truncated to the shorter one. This is the honest home of a fan-level statistical prune: pass a candidate trial’s per-seed tune column and the incumbent’s — read through Study::trial_fan — and, because Paired common-random-numbers give replicate j the same seed in both trials, element j of each column is a genuine matched pair, which is precisely what the signed-rank test assumes.

With fewer than min_pairs differences it returns false (never enough evidence to prune).