Tool and output checks
Evaluation can examine declared permissions, selected tools, structured output, and edge cases.
Test Lab
Test behavior before trust.
Compare evidence.
Prepare repeatable evaluation for activation, instruction adherence, tool selection, output structure, policy behavior, compatibility, and cross-harness differences.
Product direction · In active development
01
Test the right behavior
Planned suites cover whether capabilities activate when needed, stay quiet when they should, follow required steps, and avoid prohibited behavior.
02
Compare targets carefully
The same capability can be measured across supported harnesses for adherence, success, policy violations, latency, token use, estimated cost, and errors.
Differences are data
The goal is to reveal target-specific behavior, not flatten it into a compatibility claim.
03
Simulate before activation
Planned simulations explore configuration, skill, connector, model, policy, and rollout changes without activating them.
Preview only
Test Lab capabilities are in active development.
