The question
Which public access, product-data and policy signals can a bounded static scanner observe on a small, deliberately selected merchant sample? This pilot validates the measurement method and its failure modes. It does not estimate the proportion of all merchants that are agent-ready.
Sampling and execution
The cohort is a convenience sample of 25 public merchant storefronts. Each store is visited once by the identified L0 scanner. Stores are processed serially, with a pause between stores; each run respects robots, limits requests and samples at most three product pages and three policy pages. Private areas, checkout, accounts and purchase operations are excluded.
How to read the denominators
A blocked or unreadable store is not a failed merchant. Access results are reported over attempted stores. Product-field findings are reported only over pages and Product objects actually read. A non-variant store is not assumed from absent variant markup. We keep transport failures, robots restrictions and observed missing fields separate.
Manual review and reproducibility
A five-report agent review checks access classification, source URLs, sample size and whether a conclusion follows from its evidence. The review includes failed or blocked runs where present. Independent human review is pending. Raw merchant reports remain private in the project artifacts; this page presents aggregate counts only. A future rerun must preserve the scanner version and cohort or explicitly label the method change.
Current measurement coverage — Research HOLD
Attempted: 25. Readable homepage reports: 1. Blocked/access-limited reports: 7. Evidence-reviewed: 5. Checked: 2026-09-28T02:09:23.660Z.
25-store convenience sample; not representative. Only one homepage was readable under the scanner limits, so no product-readiness rate is estimated. Access failures are not merchant deficiencies. Five reports received an agent evidence review; independent human review is pending. Detailed robots and transport-reason totals are withheld because v1.0.0 classification issues were found and corrected in v1.0.1.