Decide: Keep, Edit, Verify, Ask Again, or Reject
Unit ID: PAW-M08-U04 Estimated active time: 20-25 minutes
Review must end in a decision
A review that ends in an impression is not finished. There are five possible endings, and naming which one you chose is what makes the workflow real.
- Keep — use as it is.
- Edit — use with your changes.
- Verify — usable once specific facts are checked.
- Ask again — the request needs fixing, not the output.
- Reject — this task is not working; do it another way.
Choose by what is wrong
DECISION RULE
What is wrong Decision
------------------------------------ -------------
nothing material keep
tone, length, a few words edit
facts you can check quickly verify
shape or emphasis is wrong throughout ask again
it invented material facts ask again, with the source bounded
you cannot tell whether it is right reject
the task needs judgement you own rejectThe last two rows are the ones people avoid, and they are the ones that matter most.
Reject is a legitimate outcome
Rejecting is not failure. It is the correct decision when you cannot verify the output, when the task carries accountability you hold, or when checking would take longer than doing it.
Teams that never reject anything are not reviewing; they are approving.
Ask again beats heavy editing
If you are editing more than about a third of an output, you are doing the work twice — once reading it, once rewriting it.
That is an ask-again, and usually a prompt problem. Note what you had to change; that is the missing constraint.
Record the decision
One line: which decision, who made it, what was checked.
For anything above the middle of the review-level table in unit 08.00, this record is what makes the step auditable.
Mini practice
Take five recent outputs and assign each one of the five decisions.
Count how many were "edit".
If most were, look at what you edited — the pattern is your next prompt improvement.