Review Evaluation Metric

A metric is a measurement used to evaluate how well a model, workflow, or system is performing.

An evaluation metric is a defined measurement used to judge some aspect of system performance, such as accuracy, error, latency, cost, safety, or user success.

Metrics turn a goal into something observable, but every metric emphasizes some outcomes and ignores others. Teams need measures that match the real use case and its harms.

A support classifier is evaluated with recall for urgent cases, false-positive rate, response latency, and escalation outcomes rather than accuracy alone.

A high metric score does not prove that a system is useful, safe, or fair. A metric can be valid for one population or task and misleading for another.

A metric measures a slice of performance, not the whole system.

* [[evaluation-gap|Evaluation Gap]] * [[verification|Verification]] * [[model-card|Model Card]] * [[calibration|Calibration]]

* [Google for Developers, “Machine Learning Glossary”](https://developers.google.com/machine-learning/glossary) — provides established boundaries for core machine-learning vocabulary. * [NIST, “The Language of Trustworthy AI”](https://doi.org/10.6028/NIST.AI.100-3) — provides a use-case-neutral vocabulary foundation for AI and machine learning.

* Compare [[evaluation-gap|Evaluation Gap]], [[verification|Verification]], [[model-card|Model Card]], and [[calibration|Calibration]] to place Evaluation Metric in the surrounding system. * Follow the cited source material above for the method, standard, or research context behind this draft.