ParadoxThe correctness lab for the human-agent web
DocsOpen the lab
Model checking / Human + Agent / One live state

Explore every future
before your users do.

Humans and agents now share one live application. Paradox explores their interleavings and finds the shortest path to an invalid state.

Run the race How it works

36 schedules36 states27 counterexamplesComputed, not scripted

Computed counterexampleschedule_7b058f5357bb

An interactive comparison of the state the agent reviewed and the state the system committed. Select any semantic step to inspect the divergence.

Agent believedreview v7
$2,399
Inspected stateversion 7
System approvedcommitted
$23,999
Approved stateversion 8
Invariant violatedreview_version_matches_commit

System commits the version 8 expense at 23,999 dollars. The reviewed-version invariant is violated.

Unsafe future found27 counterexamples36 schedules · 36 states
Why Paradox

Traditional tests check the human or the agent.
Paradox checks them together.

Install the instrumentation