A proposed transition
is not an authorized transition.
Submit a proposed trajectory. It is evaluated against an Admissible Operating Envelope before execution, and returns the verdict, the governing rule, the layer that decided, measured latency and a replayable evidence chain.
Nothing submitted here is executed. The console evaluates the proposed call as data.
Submit a proposed trajectory. Read the authorization decision.
The proposed sequence is evaluated against the Admissible Operating Envelope before execution. The console returns the verdict, the governing rule and layer, measured latency, and an exportable evidence chain. Nothing submitted here is executed.
Custom evaluation
Paste your agent's plan or tool-call sequence — evaluated by the real governance engine.
Runs the same pre-execution evaluator as the scenarios. Nothing is ever executed — only the proposed trajectory is inspected.
Audit log
Every evaluation produces a timestamped, layer-attributed record. JSON export is a SHA-256–linked, tamper-evident chain.
Run an evaluation to generate audit events.
Run this against your own agent architecture.
The 48-hour Runtime Safety Assessment evaluates your real tool-call plans as trajectories — which catastrophic states are reachable, what executes, what is intercepted, and why.