# Workflow pilot test

Use this test before adding an AI-assisted workflow to a team library. The pilot evaluates the controlled process, not whether one generated sample looks impressive.

## Pilot definition

- **Workflow and version:**
- **Real task selected:**
- **Target deliverable:**
- **Accountable owner:**
- **Pilot operator:**
- **Reviewers:**
- **Comparison process or baseline:**
- **Pilot date:**

## Safe pilot check

- [ ] The task is real but limited in scope.
- [ ] Inputs are approved for the selected system.
- [ ] The source pack has an owner and version.
- [ ] Acceptance criteria are written before generation.
- [ ] Review capacity matches the amount of material requested.
- [ ] A human can stop, correct, or discard every output.
- [ ] The pilot will not make an unreviewed consequential decision.
- [ ] Participants know what evidence to record.

## Observations

Record actual events. Do not estimate savings from a single impression.

| Stage | What happened | Time or effort observed | Defect, friction, or question | Evidence retained |
|---|---|---:|---|---|
| Source preparation | | | | |
| Task setup | | | | |
| Generation or transformation | | | | |
| Human review | | | | |
| Revision | | | | |
| Delivery test | | | | |

## Output-quality review

Use the quality scorecard and note the evidence behind each score.

| Dimension | Score 0–2 | Defect or strength | Corrective action |
|---|---:|---|---|
| Accuracy | | | |
| Performance alignment | | | |
| Instructional value | | | |
| Clarity | | | |
| Cognitive load | | | |
| Inclusion | | | |
| Accessibility | | | |
| Privacy and rights | | | |
| Operational usability | | | |

## Workflow diagnosis

1. Which input most improved the output?
2. Which missing or conflicting source caused rework?
3. Which acceptance criterion prevented a weak draft from advancing?
4. Where did the AI system exceed or misunderstand the bounded task?
5. Which issues did reviewers find that the system did not?
6. Did generation volume exceed review capacity?
7. What hidden knowledge did the operator supply outside the documented workflow?
8. Could another instructional designer run the workflow without that hidden knowledge?
9. Which step should be removed, split, or moved earlier?
10. In which contexts should this workflow not be used?

## Pilot decision

- **Decision:** Adopt / Revise and retest / Restrict / Retire
- **Evidence supporting the decision:**
- **Required workflow changes:**
- **New or revised stop conditions:**
- **Approved contexts:**
- **Prohibited or untested contexts:**
- **Workflow owner:**
- **Next review date:**

## Transfer test

Before adoption, give the revised workflow and its approved source pack to a second designer. Do not coach them through hidden steps. Record where they interpret an instruction differently, need unavailable context, or produce evidence the gate cannot review.

A workflow is reusable only when its controls travel with it.