Skip to main content
Always test before you publish. A test run shows you exactly what reviewers will see.

​
Run a test

  1. Pick a client (workstream) with real files in its Doc Store to test against.
  2. Run the agent on that client.
  3. Inspect the prepared output — the workpaper, reconciliation set, or checklist — and the sources it cited.

​
What to check

  • Completeness — did it produce everything the task should?
  • Accuracy — are the numbers and conclusions right?
  • Citations — do the cited sources actually support the output?
  • Instructions honored — did it follow any specific direction you gave (format, memo, standards)?
A common gotcha: no data to test on. Testing needs files in place. If the client’s Doc Store is empty, the test can’t run meaningfully — set up the data source first (see Connecting a data source). The guided flow points you back to the data step rather than hitting a dead end.

​
Iterate

If the output isn’t right, adjust the instructions (that fixes most behavior issues), confirm the skills and access are correct, then re-test. When it looks right, move on to Schedule and publish.

​
Next steps

  1. Schedule and publish
  2. Instructions and skills
  3. When a run fails