eproc

Anytime-valid sequential testing and confidence sequences.
git clone git://git.ppad.tech/eproc.git
Log | Files | Refs | README | LICENSE

CHANGELOG (1691B)

      1 # Changelog
      2 
      3 - 0.4.1 (2026-07-04)
      4   * Fixes a bug that made confidence sequences needlessly conservative.
      5 
      6 - 0.4.0 (2026-07-03)
      7   * Adds calibrated evidence accessors to every test module:
      8     'log_evalue', 'log_evalue_sup', and the anytime-valid 'p_value'.
      9   * Adds Numeric.Eproc.Mixture: uniform convex mixtures of
     10     e-processes, for testing a null against a union of qualitatively
     11     different alternatives at a single Ville threshold.
     12   * Adds Numeric.Eproc.ConfSeq: anytime-valid confidence sequences
     13     for bounded means, via the hedged-capital construction of
     14     Waudby-Smith & Ramdas (2024).
     15   * Adds InvalidComponentCount and InvalidGridSize to ConfigError.
     16 
     17 - 0.3.0 (2026-07-02)
     18   * Introduces a breaking API change: 'log_wealth' now returns the
     19     current log-wealth, whereas the supremum-thus-far statistic is
     20     exposed as 'log_wealth_sup'.
     21 
     22 - 0.2.2 (2026-07-02)
     23   * Adds a Numeric.Eproc.Bernoulli.TwoSided module for a two-sided
     24     Bernoulli rate test.
     25 
     26 - 0.2.1 (2026-07-02)
     27   * Two-sided bounded-mean tests now reject faster, or at least never
     28     later.
     29 
     30 - 0.2.0 (2026-06-28)
     31   * Introduces several breaking API changes, along with various internal
     32     refinements:
     33 
     34     * config functions now return 'Either ConfigError Config' and reject invalid
     35       parameters.
     36 
     37     * the 'Bernoulli.config' argument order changed from (a, p, b) to (p, a, b),
     38       for consistency with other modules.
     39 
     40     * log_wealth now returns the supremum, not the current value.
     41 
     42 - 0.1.0 (2026-06-03)
     43   * Initial release, supporting anytime-valid sequential testing via
     44     e-processes: bounded-mean, Bernoulli, and paired two-sample tests,
     45     with fixed-lambda, aGRAPA, and ONS bettors.