# AI assistant decision worksheet

Use one copy per candidate. Compare observed behavior, not a category label.

1. Job and frequency:
2. Candidate and setup (product, plan/version, test date, settings):
3. Sources and output (sample inputs, accounts, expected artifact):
4. Context and trigger (what persists; what starts the next run):
5. Allowed actions (may read/change; must never do):
6. Approval and stop (reviewer, approval boundary, cancellation):
7. Acceptance checks (three source-based facts):
   -
   -
   -
8. Failure check (missing fact/file; observed response):
9. Evidence and effort (artifact/activity references, corrections, setup/review time; mark untested):
10. Decision (limited use / retest / simpler approach; next test and reviewer):

An unresolved permission or approval failure cannot be offset by polished output.
Examples in the guide are illustrative, not comparative product-test results.

Guide: https://brainmox.com/blog/agent-vs-chatbot-vs-copilot/
Detailed scorecard: https://brainmox.com/blog/ai-assistant-evaluation-scorecard/