Skip to content
TestingModesModesAI / EvalsLLM-as-Judge

LLM-as-Judge

What it is

Claude grades soft outputs against a rubric. Verdict pass/fail plus rationale.

When to use it

Style, voice, tone — invariants that don't reduce to substring checks.

Example gates

voice-judge — shadow mode for the no-comparison voice rule.

See also

On this page