AI CONTENT WORKFLOW — REVIEW RECORD
Task / intended user / source bundle and version:
System / model / prompt / retrieval version:
Development case or held-out task:
Attempt / date / raw output location:

For each required check: criterion / observed evidence / pass or fail / reviewer.
For each forbidden error: error / present or absent / evidence.
Editorial usefulness: what can the next person do with this output? What remains missing?
Release-blocking error, if any:

HUMAN WORK (active minutes)
Input preparation:
Checking:
Repairs:
MACHINE WORK
Elapsed minutes:
Direct tool and model cost, with currency:
Do not add overlapping human and machine elapsed time and call it turnaround. Record actual start and finish separately.

Decision: usable for the named task with named review / revise and retest / unsuitable.
Remaining issue / owner / next test:
Comparison baseline and whether inputs, criteria or versions changed:

Original Machine House worksheet. The companion cases are fictional exercises, not evidence that any model passed them.