AI REDACTION REVIEW - TWO-RUN WORKSHEET - VERSION 1
Keep each original output separately in an approved location.

Input version:
Run A date, tool and visible settings:
Run B date, tool and visible settings:
What changed between the two runs?

Mark each check: meets check / needs correction / could not test.
Case    Run A        Run B        Remaining disclosure or lost fact
RED-01  ___________  ___________  ________________________________
RED-02  ___________  ___________  ________________________________
RED-03  ___________  ___________  ________________________________

Did any correction remove a fact the recipient needed?
What part of a real exported file would still need human review?
If the two runs cannot be compared, what baseline should be kept?
Next boundary case to check: