# Evaluation guide

**Authority and Evidence Layer for Physical AI Actions**

Compiled from the current public website description.
Public description compiled: 11 October 2026 · Applicability requires an environment-specific review.

## Four things to evaluate

1. Check that an unpermitted command is not forwarded through the integrated path
2. Check that an authorized request reaches the test output
3. Check that a request outside defined bounds is refused with a reason
4. Check record authenticity and integrity using the supplied verification material

These are evaluation cases, not results of an execution, physical safety test or safety certification.

## Suggested starting point

Bring the robot model, command interface, action to check and existing safety setup to define an evaluation scope together.
Start by comparing the ROS 2 reference path with your actual command path. Applicability to your robot requires a separate review.
Review the command interface and existing control path before defining adapter integration scope. Immediate compatibility is not implied.

Public description files are not authorization tokens or robot-control interfaces.

Contact: ygkang@glowfly.ai
Source: https://www.glowfly.ai/en/