What to inspect
Review the sequence of inputs, retrieval, tool selection, tool arguments, tool result and final synthesis. Locate the first divergence from expected behavior. Correcting the final wording does not address a bad record lookup earlier in the trace.
Worked example
A synthetic trace retrieves customer C-140 for a request about C-104. The tool succeeds and the final answer accurately summarizes the wrong customer. The earliest failing step is record selection, not final text generation.
Copy the full prompt
Adapt the inputs to your task. Remove private data before sending anything to a model. This page copies text locally; it does not run the prompt.
Review [REDACTED TRACE] against [EXPECTED WORKFLOW]. For each step record input, selected action, output and expected behavior. Identify the earliest observable divergence. Separate evidence from speculation. Propose one targeted regression test and identify any missing trace fields. Do not infer hidden reasoning or expose secrets from logs.
Review checklist
- Redact secrets and personal data before sharing traces.
- Preserve step and call identifiers.
- Find the earliest divergence.
- Test the cause instead of only rewriting the last answer.
Keep the resource
Download the complete note ↓Markdown · explanation, example, prompt and checklistDownload the prompt ↓Plain text · ready to adaptOpen the full-size visual ↓SVG · scalable reference diagramDocumentation & scope
OpenAI: evaluation best practices ↗
This is an original workflow template with an illustrative example. It does not report completed model tests or measured performance. Provider APIs can change; check the linked documentation for your exact integration.