Add Review Gates, Privacy Checks, and Stop Conditions

Unit ID: PAW-M12-U04 Estimated active time: 25-30 minutes

A good prompt is not a finished workflow

By this point the prompt works. What is missing is everything around it: who checks, what stops it, what is recorded.

Module 11 built these separately. This unit attaches them to your capstone task.

Set the review level from consequence

Look up your task on the review-level table from unit 08.00, and be honest about the consequence row.

Most people place their own task one row too low, because they are thinking about the usual case rather than the bad one.

The capstone gate sheet

GATES — (task name)

PRIVACY GATE
  Categories that never go in:
  Substitutions used:
  Re-identification risk checked by:

REVIEW GATE
  Reviewer (not the prompt author):
  Against:                     rubric v__
  Time budget:
  Decision recorded as:        keep / edit / verify / ask again / reject

STOP CONDITIONS  (three to seven)
  1.                           -> instead:
  2.                           -> instead:
  3.                           -> instead:

APPROVAL
  Named approver:
  Confirms:
  Record kept:                 where, how long

Test the gates before you finish

Walk one real instance through, gate by gate.

Time each gate. If any takes more than a minute to answer, simplify it — a gate nobody has time for is not a gate.

Write the failure path

What happens when the tool is unavailable, when the output fails review twice, or when a stop condition fires.

A workflow without a manual path stops working the day something changes, and that is exactly when it matters.

Mini practice

Complete the gate sheet for your task.

Walk one real instance through and time each gate.

Write the manual path in two sentences.

Mark complete