How it works
Category methodA bounded path from Human intent to Human acceptance.
VerShep is designed to preserve the decision path around consequential AI-assisted work, even when the underlying model or tool changes.
- 01
Ground the request
Resolve the intended outcome and authoritative sources.
- 02
Bound the authority
Set permissions, non-goals, budgets, and stop conditions.
- 03
Verify the result
Run independent checks and record what remains unknown.
- 04
Accept the outcome
Return the evidence to a named Human for the final decision.
Stable assurance
The framework stays consistent while the runtime changes.
May change
Models, agents, tools, providers, repositories, and environments.
Must remain visible
Source authority, delegated purpose, action boundary, independent evidence, open findings, and Human decision.
Why a shepherd
Guidance without replacing judgment.
VerShep, the Border Collie, makes the behavior of the system visible: understand the Human direction, maintain boundaries, detect drift, gather evidence, and return the work for a Human decision.
The robotic sheep make the agent relationship visible. Healthy agents continue through the bounded course; suspicious agents are restricted, isolated, and presented for the appropriate response.
VerShep never approves an outcome and never represents autonomous Human authority.