# Mastra Evals Evals (and scorers) are [[Mastra AI]]'s quantitative quality-measurement system for agent and workflow outputs. A scorer returns a numerical score (typically 0–1) so you can track quality over time and run comparative experiments. ## Key aspects - **Live evaluations**: real-time, asynchronous scoring of production outputs - **Trace evaluations**: historical analysis over recorded [[Mastra Observability|traces]] - Scorers track trends and power A/B experiments through [[Mastra Studio]] - Custom and built-in scorers ## References - https://mastra.ai/docs/evals/overview ## Related - [[Mastra AI]] - [[Mastra Agents]] - [[Mastra Observability]] - [[Mastra Studio]]