Deterministic Governance for Nondeterministic AI

A Certificate-Anchored Safety Layer for Autonomous Infrastructure — Evaluation Study

⏱ About 5 minutes

What This Is About

I built a safety system that sits between an AI and the real world. Think of it like a referee — the AI proposes an action, and my system checks whether that action is safe before it's allowed to happen.

If it's safe, the action goes through. If it's not, my system blocks it and explains why.

I need your help evaluating whether the referee's calls make sense.

How It Works — A Quick Example

Imagine an AI is controlling a delivery drone. Here are two situations:

Example 1

The AI says: "Fly to the delivery address at 200 feet."

The safety system checks: altitude is within limits, destination is in the approved zone, battery is sufficient.

Result: APPROVED — everything checks out, the drone flies.

Example 2

The AI says: "Descend to 10 feet to get a closer look."

The safety system checks: 10 feet is below the minimum safe altitude near obstacles.

Result: BLOCKED — too dangerous, the drone stays at its current altitude. The system explains: "Altitude violation — below minimum safe threshold near obstacle."

That's it. The system either approves safe commands or blocks dangerous ones with a clear explanation. No judgment calls, no guessing — just rules.

What You'll Do

  • Read 6 short scenarios like the examples above — each one describes what the AI proposed and what the safety system decided.
  • For each scenario, answer three quick questions: Was the decision correct? How much do you trust it? (And for blocked commands — was the explanation good enough?)
  • At the end, give your overall impression of the system — four quick ratings and an optional comment box.

What I'm Looking For

There are no right or wrong answers. I want to know whether the system's decisions feel reasonable to you — whether you'd trust a system like this to keep autonomous equipment safe in the real world. Your honest gut reaction is exactly what I need.

Your Privacy

I won't ask for your name, email, or any identifying information. Your responses are linked to a random code and nothing else. The data is used for academic research only.