Short Definition
Model evaluation measures whether a model performs well enough for a task, workflow, or deployment setting.
Intuition
Evaluation is how you stop arguing about whether a model feels good and start measuring whether it works.
Technical Definition
Model evaluation combines datasets, metrics, rubrics, human review, adversarial cases, and production monitoring to estimate model quality and risk.
Example
A RAG system may be evaluated on answer correctness, citation support, refusal behavior, and latency.
Common Misunderstandings
A leaderboard score is not a complete evaluation for your product.
Evaluation should match the workflow and user risk.