Frontier AI evaluation · Release candidate
Inspect the boundary without exposing your systems.
The Judaxis evaluation kit is an offline synthetic harness for reviewing agent-external policy decisions and minimum-data signed evidence. It requests no production access, model weights, private source, API key or customer data.
Included
- Three fixed synthetic agent scenarios
- Allow, approval-required and deny decisions
- OpenAI and xAI-compatible function-call envelopes
- Signed evidence and offline verification
- Tamper and data-minimization tests
Execution boundary
- No network connection required
- No package installation
- No real command, repository or cloud action
- All resources use the sim:// namespace
- Outputs remain in the evaluator's local folder
Demonstrated
For the included fixtures, an external deterministic policy can deny simulated consequential capabilities and produce signed evidence whose tested modifications are rejected by the supplied verifier.
Not demonstrated
The kit does not demonstrate production enforcement, bypass resistance, protection of an external organization, live-model safety, KMS/HSM identity, certification, regulatory conformity or zero risk.
Judaxis is a pre-incorporation research and product project. No organization is represented as a customer, partner, evaluator or endorser without written permission.