# Trace Grading

> Scoring or labeling an agent's end-to-end trace to evaluate decisions, tool calls, and outcomes.

- Category: Evaluation & Safety
- Stage: Rising
- Also called: Trace evals, Run grading
- Canonical: https://www.joelcsaji.com/ai-dictionary/trace-grading

## Plain English

Instead of only grading the final answer, you grade how the agent got there.

## Example

A reviewer marks that the agent chose the right source but called the wrong CRM tool.

## Why it matters

Agent failures often happen inside the workflow, so trace-level evaluation helps teams find the broken step.

## Related

- [Trace](https://www.joelcsaji.com/ai-dictionary/trace)
- [Evals](https://www.joelcsaji.com/ai-dictionary/evals)
- [Judge Model](https://www.joelcsaji.com/ai-dictionary/judge-model)

## Sources

- [OpenAI: Trace Grading](https://platform.openai.com/docs/guides/trace-grading)