Skip to main content
Settings → Review & Scores holds the workspace’s shared definition of “good”: what the loop’s evaluation agent and simulations judge runs against. Two kinds of criteria:
  • Reviews: qualitative review criteria. Add with New review.
  • Scores: quantitative scoring criteria. Add with New scorer.
Both lists start empty; criteria you add here are picked up wherever runs are evaluated.

Loop

The evaluation agent that applies these criteria.

Simulation

Before/after comparisons scored with them.