Skip to content

Roadmap

What is planned for the next releases of evalproof. It is for people deciding whether a feature they need is coming. The plan is subject to change: an item can move, shrink or be dropped, and nothing here is a commitment to a date.

What has already shipped is in the changelog. Version 1.0 declared the public surface stable (Versioning and support); new features now arrive in minor releases without breaking what works.

Later

  • A coding-agent add-on. Suites for coding agents: tasks that run against a repository, and checks that read whether the tests pass.
  • Stratified gold sampling and per-tag calibration. Draw the gold sample so that each tag is represented, and calibrate the judge per tag rather than over the whole run.
  • A published study comparing models on an agent suite.

Next steps