A benchmark for high-dimensional biomedical tables.

Primary analysis: each score comes from the requested feature- and sample-cap cell; no smaller cell is substituted. Tuned failures use the untuned parent; other failures use the constant predictor.Sensitivity analysis: after a memory failure, a verified result from a smaller sample cap may be carried forward. It is not performance at the requested larger cell.Sensitivity analysis: a failed run is omitted instead of imputed, which removes that model from the target. Models are then ranked on different target pools, so this view is an optimistic bound rather than a like-for-like comparison.