+ Evaluation / Governance / Assurance
Explore the system.
Follow the failure.
Twelve interactive studies. Change a condition, inspect the response and follow the finding through to a retest.
A clear path.
A hidden assumption.
A robot plans a route from a sensor reading. What happens when that reading is stale, but the movement command still looks valid?
The location estimate is recent enough to authorize movement.
Introduce a disruption to expose the dependency behind the action.
A conceptual model, not a measured safety score. Change the pressure and compare the response.
One failure can
reach another system.
A robot needs sensors. A sensor needs a clock. A remote operator needs a network. We test what happens when those dependencies fail.
AI agents that use tools.
Test how agents use tools, memory, and delegated permissions.
Retrieved content → delegated authority → external action
- Test failures and record what happened
- Set permissions and name decision owners
- Repeat tests when the system changes
Illustrative dependencies. Each engagement starts with the actual system.
Find the fault.
Check the fix.
We test the system, document what happened, identify who needs to act, and test the response.
What needs to be true
before the system acts?
Agree what the system is meant to do, who could be affected, and what the test must show.
START WITH A QUESTION
Research Triangle Park, North Carolina.
Working across systems. Working globally.