# Human Approval Gate facilitator notes

Status: published Version 1.0.2

Published: 2026-08-28

License: Creative Commons Attribution 4.0 International (CC BY 4.0)  
https://creativecommons.org/licenses/by/4.0/

## Session

Title: Make Human Approval a Real Workflow Step

Length: 30 minutes

Audience: leaders, educators, operators, and small teams using AI-assisted workflows

Learning objective: participants can define and test a human approval gate using the CLEAR test without using private workplace information.

## Safety boundary

Use only the synthetic workshop-announcement example or another invented scenario. Ask participants not to share patient, employee, employer, client, security, or confidential information. This is an educational practice session, not legal, clinical, privacy, or security advice.

## Materials

- practitioner guide
- one worksheet per participant
- ten synthetic test cases
- timer
- optional shared screen with no private data

## Run of show

### 0 to 4 minutes: define the problem

Ask: "If a workflow says a human reviews the output, what exactly stops the next action?"

Define a human approval gate as a controlled decision point where a named, qualified person checks defined evidence against written criteria before consequence.

### 4 to 10 minutes: teach CLEAR

- Consequence held
- Listed reviewer
- Evidence checked
- Allowed outcomes
- Record and recovery

Stress that a polished output is not evidence of correctness.

### 10 to 20 minutes: complete the worksheet

Use the synthetic public-workshop announcement. Participants name the held consequence, reviewer, evidence, criteria, outcomes, and recovery path.

### 20 to 27 minutes: test the gate

Choose three cases:

1. private-data marker appears in the input
2. the registration link conflicts with the event brief
3. the text changes after approval

For each case, ask whether the workflow accepts, revises, rejects, or escalates and which evidence supports that decision.

### 27 to 30 minutes: close

Ask each participant to state one observable control they would require before a real workflow acts. Remind them to take the method back through their own policy, privacy, security, legal, and professional review.

## Debrief questions

- Was the consequence truly held?
- Could the named reviewer reach a decision from the evidence?
- Were the criteria written before the output appeared?
- Could the reviewer reject without being overridden?
- Did approval remain attached to the exact version used?
- Could the team recover without a duplicate action?

## Evidence to preserve

For an authorized public workshop, preserve only attendance counts, completed synthetic worksheets if participants consent, facilitator observations, and anonymous improvement notes. Do not collect workplace examples or imply that attendance validates the method.