2.0
Datasets
Open datasets and benchmarks for AI agents that shop, compare and pay, starting with CartTrace-100k, our first release. Alongside them, retail shelf data from real stores for the models that run inside them.
CartTrace-100k
101,416 journeys of an AI shopping agent through checkout and payment over UCP, ACP, AP2, MPP, Shop Pay and WebMCP, across 1,044 stand-in stores in 25 categories, each scored by a judge. Our first public release, open on Hugging Face.
Commerce Agent Traces
End-to-end recordings of AI agents navigating e-commerce: product search, comparison, cart, and checkout flows across live retail environments. Built by Machine Commerce for training and evaluating commercial agents. In development.
CartBench
A structured evaluation suite for commerce agents, built from Machine Commerce behavioral traces. Covers product discovery, price comparison, cart management, and end-to-end checkout, scored for task completion, step efficiency, and robustness to real-world failure modes. In design; planned for early 2027.
ShelfTrace
Real shelves in real stores, where every product box records where it sits, what it is, whether it is real, who decided, and what it sold. The training and evaluation set behind Shelf-1. Private; partnership access.