Ranks of the across-repetition mean performance at n_train = 121,626 across the reported metrics and evaluation regimes. Rank 1 is best within each column; coverage is ranked by distance from 0.90, and bias is ranked by absolute value. White cells denote metrics that do not apply to point-only estimators. Ranks are descriptive and do not account for uncertainty; inferential comparisons therefore use the standard errors and bold sets in the main tables. The raw α = 1 regime has an effective sample size of only 216 objects (1.6%) and should be interpreted cautiously.