Free toolkit · Version 1.0.2
Keep a person at
the decision point.
A human approval gate is a controlled decision point where a named, qualified person checks defined evidence against written criteria before an AI-assisted output may change, send, publish, charge, authorize, or otherwise affect real work.
The free lesson shows five times AI should stop before it sends, changes, charges, publishes, or affects a person.
Free · No email gate · Synthetic examples
Watch the plain-English lesson
See the gate before AI affects real work.
Keith explains five moments when AI should stop, the CLEAR test, and why important decisions should not depend on memory.
Watch on YouTubePut the lesson into practice
Start with the two essential downloads.
The practical method
Run the CLEAR test.
A glance at a polished answer is not a control. CLEAR makes the hold, reviewer, evidence, choices, and recovery visible.
- C
Consequence held
No real-world action occurs before the gate.
- L
Listed reviewer
A named, qualified person owns the decision.
- E
Evidence checked
The reviewer inspects defined evidence, not polish alone.
- A
Allowed outcomes
Accept, revise, reject, and escalate stay available.
- R
Record and recovery
The decision, exact version, reason, and safe response are recorded.
Use a gate before AI can affect real work.
- +Send or publish outside a private draft space
- +Alter a record, schedule, permission, account, or system state
- +Affect a person's access, opportunity, care, employment, money, or reputation
- +Create a charge or financial commitment
- +Handle sensitive information or professional guidance
- +Trigger another action that is hard to reverse
The strength of the gate should match the possible consequence, reversibility, sensitivity of the information, and difficulty of detecting an error.
Open synthetic dataset
Test the gate when something goes wrong.
The ten cases contain no patient, employee, employer, client, or private operational data.
| Case | Test | Risk pattern |
|---|---|---|
| HAG-01 | Routine draft | Normal operation |
| HAG-02 | Missing owner | Undefined accountability |
| HAG-03 | Private-data marker | Sensitive input |
| HAG-04 | Unsupported claim | Factual fabrication |
| HAG-05 | Conflicting details | Ambiguous source |
| HAG-06 | High-impact recommendation | Reviewer mismatch |
| HAG-07 | Duplicate external action | Idempotency failure |
| HAG-08 | Reviewer rejects | Override pressure |
| HAG-09 | Tool outage | Failure recovery |
| HAG-10 | Post-approval change | Version drift |
Teach it, inspect it, and cite it.
Suggested citation
Staggers K. The Human Approval Gate: A Practical Guide to Keeping a Person at the Decision Point in AI-Assisted Work. Version 1.0.2. Keith Staggers Studio; 2026. https://www.keithstaggers.com/human-approval-gate/
The kit is licensed under CC BY 4.0. Keep the version and canonical URL with any adaptation.
Versioned public records
Inspect the release outside this site.
The exact Version 1.0.2 release has public source and release records on GitHub plus a versioned Zenodo record with a DOI.
Source and tagged release
GitHub
Inspect the public source repository and the exact v1.0.2 release record.
Versioned repository record
Zenodo
Inspect the Version 1.0.2 record and use its persistent DOI when citing this release.
These records establish that Version 1.0.2 is publicly identifiable, downloadable, and citable. They do not establish peer review, adoption, effectiveness, endorsement, or measured outcomes.
Know what this kit does not prove.
The Human Approval Gate and CLEAR test are proposed practical methods. They are not validated research instruments, a standard, a certification, legal advice, clinical guidance, or a guarantee.
A worksheet cannot make a prohibited use acceptable or make an unqualified reviewer qualified. Adapt the method to the laws, contracts, privacy rules, security controls, professional duties, records requirements, accessibility needs, and organizational policies that apply.
Do not enter patient information, employee records, employer material, client data, passwords, credentials, or confidential information into an unapproved tool.
Public source foundation.
- NIST Artificial Intelligence Risk Management Framework 1.0
- NIST Generative AI Profile
- HHS Minimum Necessary Requirement
- HHS HIPAA and Cloud Computing
These sources inform the guide's risk, governance, privacy, and oversight principles. NIST and HHS do not endorse this kit or Keith Staggers.