Writing

How it was built,
and what it failed to find.

Essays and records from building flakestat.

  1. Essay8 September 2026

    Validating a flaky-test detector

    A week to build it. Rather longer to work out whether it did anything, and three bugs that all had the same shape.

    Read
  2. Validation record8 September 2026

    How flakestat was validated

    Three third-party subjects, one frozen build, and a set of rules fixed in advance about what each result would be allowed to mean.

    Read
  3. Field notes8 September 2026

    Hunting flaky tests in open source

    An overnight run against five active Go repositories, and why it closed nothing.

    Read
  4. Design8 September 2026

    Design notes

    Why flakiness is scored as inconsistency, and the three times that principle was implemented wrong.

    Read
  5. Releases8 September 2026

    Changelog

    Every release, and which changes move scores you have already recorded.

    Read