For the complete documentation index, see llms.txt. This page is also available as Markdown.

GEPA iterative labeling

Recipe — iteratively label examples and re-optimize judges as your bar evolves.

Treat judge optimization as a continuous loop:

  1. Use judge in production / pre-deploy

  2. When verdicts disagree with reviewers, log + label

  3. Quarterly: append new labels to the training set; re-optimize

  4. Compare old vs new agreement rate

This keeps the judge calibrated to your team's evolving bar.

See also

Last updated

Was this helpful?