Failure Lab and live data
Select live data and Failure Lab answer different questions. The first limits what a real run may read; the second checks whether the workflow has the controls it needs before anything runs.
Select the real sources
Before a managed Live run, Select live data (Pro) shows the connected sources referenced by the team. Keep only the sources this run needs. A step that reaches an unselected integration stops with a clear setup result instead of inventing data.
- Open the Run menu and choose Select live data (Pro).
- Review the source, account, and access mode for this run.
- Start the run and use the terminal to confirm which source each step actually used.
Run Failure Lab locally
Failure Lab performs nine deterministic readiness checks for timeouts, rate limits, malformed output, authentication, prompt injection, approvals, cost caps, and partial failure. It does not call a model, connector, browser session, or billable executor.
Use each result correctly
Fix structural findings in Team Flow, rerun Failure Lab, then use Dry run to inspect routing. Move to Live run only when setup is green and the selected live-data boundary matches the job.
- Failure Lab is free static readiness analysis.
- Dry run simulates order, handoffs, and approvals.
- Live run calls the selected models and supported tools.
- History keeps the final evidence and cost receipt.