Get In, Decide, Prove: A Merchant-Side Readiness Benchmark for Agentic Commerce

Key points

• Far less is known about the other side of the transaction: whether a merchant's website lets an agent in, publishes terms precise enough to decide a purchase, and leaves evidence of what was agreed.1
• This paper presents a merchant-side benchmark built on those three questions and reports a pilot of it.1
• The central measurement, decision sufficiency, rates each criterion of a concrete buyer mandate (total cost, cancellation or return terms, service date, variant identity) as determinate, indeterminate, or unknown from the merchant's own published evidence.1
• Readiness scores carry their evidence coverage and missingness bounds; access is reported separately for a declared bot, a declared headless browser, and a user-present browser; and each observed term is mapped to the RAILS admissibility class of the evidence that could support it in a dispute.1

Sources (1)

• [1]Get In, Decide, Prove: A Merchant-Side Readiness Benchmark for Agentic Commerce
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 08:35 PM

  AI agents now browse and buy on people's behalf, and benchmarks measure how well the agents perform.

  Far less is known about the other side of the transaction: whether a merchant's website lets an agent in, publishes terms precise enough to decide a purchase, and leaves evidence of what was agreed.

Extractive summary: sentences quoted from the sources.