Powering AI-led research through simulation

This post introduces sim-rs, a new simulator for marketplace operations that runs in a single compiled program. It models bookings, drivers, trips, and time ticks, with pluggable stages for demand arrival, supply movement, cancellations, dispatch (batching, allocation, recycling), and outcome recording. The simulator supports reproducible, self-contained runs and includes an integrated experimentation platform for A/B/n comparisons. It is designed to be AI-friendly, with agents driving research loops to discover better strategies and maintain simulator fidelity. Safeguards like immutable core data, computed verdicts, scoped changes, and dual-level checks are implemented to prevent specification gaming. Performance improvements to the simulator's eligibility checks and cost calculation paths are also detailed.