RankTieR: deterministic gene-set scoring and random-tie-breaking lottery diagnostics (code and reproduction package)

RankTieR is a deterministic mean-rank rank-AUC gene-set scorer; the package is bit-for-bit reproducible on tie-dense sparse single-cell matrices. The reproduction chain includes: (1) the sparse-count benchmark grid (3 sparsity levels x 5 gene-set sizes x 3 effect conditions, 30 replicates, 5 scoring methods, AUCell 20-seed redraws); (2) the lottery-ruler formalization (per-cell redraw SD sigma_L, signal-to-noise ratio R = |delta|/sigma_L, redraw flip rate F, with the standardized delta convention); (3) the retrospective-audit scripts with per-study mini-manifests and md5 checksums.

Fixed random seed SEED0 = 20260922. Environment: R 4.5.3, AUCell 1.32.0, UCell 2.14.0. All audited datasets are public GEO resources; accession identifiers are disclosed in the manuscript supplementary materials.