Skip to content

2026 – present

stats-claw

In-process statistical computing for Rust — distributions, hypothesis tests, and resampling — with zero runtime dependencies, validated against scipy.

Data science on the hot path

0

Runtime dependencies

14

Distribution families

1e-12

Tolerance vs. scipy

Detail

  • The Rust ecosystem has strong data pipelines and inference, but no classical hypothesis-test suite with scipy-grade p-values. That gap is the crate's reason to exist.
  • std-only: no BLAS, no LAPACK, no transitive supply chain. It compiles in seconds and drops into any Rust binary, including constrained targets.
  • Fourteen distribution families with pdf/pmf, cdf, quantile, moments, log-space tails, and seeded sampling. Exact and asymptotic p-values across t, ANOVA, χ², Fisher, Mann–Whitney, Wilcoxon, Kruskal–Wallis, KS, and Shapiro.
  • Every numeric is checked against committed golden fixtures generated from the reference Python libraries, to tolerances documented per area.

Stack

RustKaniMiriscipy equivalenceSplitMix64

Underlined entries link to the writing on that topic.

Links

All projects