bonsai
← Labs
Foundations
~8 min
requires API key

Rubric builder & dry-run

Iterate on a rubric and see how scores change live.

Drop a candidate output into the editor, write criteria, and watch a Claude judge score it under each rubric variant. Designed to teach rubric craft without needing infrastructure.

Learning objectives
  • ·Translate vague quality goals into checkable criteria.
  • ·See how rubric phrasing changes scores.
  • ·Recognize when a criterion needs evidence requirements vs. a binary check.

Candidate output

Try a rubric variant

Try varying phrasings of the same idea: a vague version (“is high quality”), a specific binary version (“contains zero claims of being ‘best’, ‘world-class’, or ‘industry leading’”), and an evidence-required version. Watch how the verdict and the quoted evidence change.