# One workflow: acceptance and handoff

Job owner:
Input and source:
Expected completed outcome:

## Before the pilot
Measure the current human time on representative work.
Name the current source and its owner.
Set critical failure conditions separately from average quality scores.
Define permitted actions, reviewer and fallback process.

## Include awkward cases
Missing information.
Conflicting or outdated policy.
Out-of-scope request.
Unavailable tool or failed save.
Duplicate input.
An instruction inside a source document that conflicts with the agreed task.

## Observe every eligible item
Did an output arrive?
Did it pass before human correction?
How much review and exception work did it create?
Was the actual destination correct?
Did it eventually complete after correction?

Count missing and failed outputs. Keep first-pass acceptance separate from eventual completion. A small pilot is a learning exercise, not a statistical reliability claim.

## Decide
Keep, revise or stop:
Evidence:
Unresolved risk or failure:
Next check and owner:
Retest after model, policy, tool or workflow changes.

DLVX practical AI guides. This is a template, not a claim that a workflow has passed.