# Changelog

## 0.1.0 - 2026-06-30

- Added descriptive statistics for one-dimensional finite samples.
- Added Normal and Student's t distribution helpers.
- Added one-sample, paired, Welch, and pooled t-tests.
- Added average-rank utilities.
- Added asymptotic, exact, and auto Mann-Whitney U tests.
- Added JSONL fixtures generated from pinned NumPy, SciPy, and Statsmodels
  references.
- Added upstream-derived fixture coverage notes.
- Added `nan_policy: :raise | :propagate | :omit` for current public statistics
  APIs.
- Added t-test confidence intervals and optional Cohen's d / Hedges' g effect
  sizes.
- Added Mann-Whitney common-language and rank-biserial effect sizes.
- Added `effect_size.cliffs_delta` as an alias of Mann-Whitney rank-biserial.
- Added dataframe-style `columns:` and `pairs:` wrappers for t-tests and
  Mann-Whitney U tests, including optional `Explorer.DataFrame` support when
  Explorer is loaded by the caller.
- Added opt-in tensor extraction for dataframe columns with `input: :tensor`
  and opt-in Nx reduction paths with `backend: :tensor`.
- Added the `Statwise.Visualization` API for semantic chart construction,
  Vega-Lite export, optional Livebook/Kino display, themes, faceting, and
  composition.
- Added statistical result annotations for categorical plots, including
  t-test and Mann-Whitney comparison brackets, computed per-facet tests, and
  p-value/statistic/effect-size labels.
- Added runnable Livebook galleries for visualization features and statistical
  test selection, variants, dataframe-style inputs, and plot annotations.
- Added a Python-reference benchmark harness for comparing Statwise against
  pinned NumPy, SciPy, and Statsmodels calls.
- Added benchmark JSON output and baseline comparison support for regression
  checks.