Posts

Write-ups on environments, graders and evals.

• September 4, 2026
  Suitesmith: training a 4B model to write test suites

  An RL environment which trains a model to write test suites for functions which operate on structured data. 20 steps of GRPO took a 4B from 0.41 to 0.88, and evaluating the same checkpoint twice disagreed by 0.13 on the tier that mattered.