Freeze the synthetic task at its last complete stage
Suppose the fixed workshop task has selected its source note and produced the first brief containing the Saturday sentence. The separate review cannot start on its chosen path.
Keep the note and first brief unchanged. Mark review, correction, and checkpoint as incomplete. Do not present the first brief as the final result.
This scenario uses the same synthetic job. It demonstrates stop behavior, not service reliability or provider performance.
Explain the gap in ordinary language
The stop record needs four facts:
- Which stage finished?
- Which stage could not begin?
- What requirement is missing?
- What decision can resume the task?
It does not need private account details, system addresses, usage data, or hidden selection rules.
Avoid silent fallback
A different model may interpret sources, tone, or review criteria differently. Substitution can be reasonable, but it should be a deliberate choice recorded with the task.
If the required review perspective cannot be preserved, stop and ask for a new decision. Do not weaken the review invisibly to make the sequence appear complete.
Preserve a resumable checkpoint
Save the source identities, task definition, completed stages, unfinished stage, and known unknowns. The next run should continue from that state instead of rebuilding context from memory.
When a replacement path is chosen, apply the same review question to the same first brief. This keeps the comparison stable.
Distinguish a safe stop from a finished result
A finished result includes the corrected brief and final checkpoint export. A safe stop includes enough information to continue, but it does not claim completion.
That distinction protects the reader from treating partial work as a reviewed artifact.
Fit and poor fit
Explicit stop behavior fits multi-stage work where model choice or review role matters. A fixed transformation may instead return a direct technical error and wait for repair.