RAINLIGHT AI

AI Workflow Acceptance Testing: What to Check Before Launch

Copy this original working template and adapt it to your process. Examples are illustrative; this is not a performance guarantee.

Workflow / version / owner / date:
Eligible input and definition of completed work:
Test environment and harmless destination:
Actions allowed automatically / actions requiring approval:
For EACH test: input ID / expected result / actual result / destination receipt / pass or fail / impact / correction owner:
Cases: normal / missing data / ambiguous / duplicate / timeout / expired access / opt-out / denied approval / cancellation / untrusted instructions.
Held-out examples not used while tuning:
Correct cases / total cases by task (not one unexplained score):
Blocking failures and unresolved limitations:
No-response approval path / backup reviewer / expiry:
How uncertain external outcomes are reconciled before retry:
Pilot scope and spending limit:
How new actions are paused:
How in-flight actions and pending work are reconciled:
Human fallback and restoration procedure:
Business outcome / quality / handling cost to monitor:
Acceptance decision / decision maker / review date:
Passing this checklist is not a guarantee of zero production failures.