Skip to content

Scenario explorer

Every eval scenario, replayed through the real runtime with a fake clock and the scripted model: what came in, what the agent decided and why, what it sent, and the state of every open ask after each step. The data is produced at build time by uv run load-agent export; 30 of 30 scenarios match their expectations.

30 scenarios

Prepared for Freight Hero by Marcus Caum Source on GitHub