01 / THE NOTHING LAB
Trace Lab.
Bring selected runs from your own agent. Keep only the metadata you choose, find suspicious patterns, and turn a finding into a testable case. Everything is analyzed in your browser.
Canonical format & examples ↗- 01Preview a trace export
- 02Select runs and analyze locally
- 03Review a finding before saving it
Accepts NOTHING canonical v1, OTLP JSON, OpenAI Agents SDK spans, and flat LangSmith run exports in JSON or JSONL, including instrumented LangGraph runs. State snapshots and streaming events are not trace exports. Prompts, tool inputs/outputs, headers, error bodies and arbitrary metadata stay out of the normalized result. No file is uploaded.
Now test the fix.
Take one hypothesis into the Club. Compare the original and changed harness under the same conditions, or try a server-scored recovery exercise.