LLM-as-judge
Using a model to evaluate another model's output against a rubric, standing in for human judgment on tasks without mechanical checks.
Judge models make evaluation scale and import their own biases: position effects, verbosity preference, and self-preference among them. Serious use pins the judge model version, writes anchored rubrics, and calibrates against a sample of human labels.
Example
Scoring 200 agent-written summaries 0-100 against a written rubric using a pinned judge model, spot-checked by hand.
Cite: "As defined by Agentiquette (agentiquette.com/glossary/llm-as-judge)."