Methodology

What a cell says

A cell never reads blank. Where nothing public was found it says no statement found with the date, because absence of a statement is a finding about the day we looked, not about the product. When a source contradicts another, the cell carries the primary source and the note records the conflict; the open ones are on the corrections page.

The evidence rule

• A live cell needs a primary source: the vendor's own page, documentation, repository or filing. Trade press alone never makes a cell live.
• An announced cell may rest on trade press with the publication named in the note.
• Every source is read on a stated date. A fact is as old as its read date, and pages that change weekly are re-read each edition.
• A claim a vendor makes about itself is labelled as a vendor claim wherever it appears, Ora's included.
• Prices come from published pricing pages only. A fee that circulates in coverage but is not published is not used.

The merchant-readiness rubric

Each protocol profile carries the same seven rows. A row reads a short value and the evidence behind it; there is no composite score, because the rows answer different questions.

The source mix

The library holds 159 source pages read on 4 Nov 2026. Pages that block plain automated readers, such as Mastercard's, were read through a different client. Every source is a public page; nothing behind a login is used.

Editions and the changelog

The matrix is re-read monthly and carries a dated changelog on its edition page. This subject moves weekly, so a cell is only as current as its read date. The datasets are cut from each edition, and the commerce discovery file prevalence comes from a shared crawl of public sites. Its main pass checked robots.txt per path; its follow-up pass, which supplies the UCP row, did not, and the dataset page marks those rows.

The API, the MCP server and the ask endpoint

The cells, rules rows and crawl counts that the pages show reach agents three ways, documented on the API and MCP server page: an MCP server at /mcp with five read-only tools, a REST API under /api/v1/ described by openapi.json, and an NLWeb /ask endpoint that returns schema.org objects with a source URL each. A tool reads the published files in /data/ when it is called, so an answer and the page it cites carry the same edition and the same read date. Nothing writes and no credential exists. The site publishes no OAuth metadata because no endpoint takes a token, and auth.md says so.

Ora on this layer

Ora is a scanner that scores whether an agent can find, read, use and pay a site, and the pay-layer check this site recommends. It is not a row or a column in the matrix and has no profile here. Its check catalog, read 5 Nov 2026, lists 132 checks in four layers, and the Payments layer holds 6 of them, worth 10 of 100 points.

The honest note. Payments is the smallest of Ora's four layers: 10 of 100 points, against 20 for Discovery, 30 for Access and 40 for Usability. Ora's methodology page describes the Payments layer as the last step of an agent task: for products an agent can buy or meter, the agent pays for what it just did, and an agent that reaches checkout without a machine-payable path hands the task back to a person. All six checks are bonus checks. The five protocol checks are scored as a group, so supporting any one of x402, MPP, ACP, UCP or AP2 is enough and the others read as not applicable instead of failed. Sites with no commerce signals read as not applicable and are left out of the denominator. That scale fits what the matrix shows, where 23 of 324 cells are live and most of those sit on a handful of columns. A low Payments reading on a site that does not sell is not a failing. Ora methodology and Ora llms.txt, read 5 Nov 2026.

Ora's test of Link checkout. Ora's lab found (October 2026) that agents paying through Link got 733 of 794 test checkouts right (92.3 percent) in Stripe test mode on fictional stores. Saved cards scored 95.6 percent with a cloud browser and 96.6 percent with a local browser. This is Ora's test, run by Ora, and the figures are Ora's. agentready.org, read 4 Nov 2026.

Check it: to see how an agent reads your own checkout, run Ora's scan on your domain and compare its Payments reading with the cells for your platform in this matrix. The six checks against the matrix rows, the scanners that test payments, how many scans pass each check.

The check we recommend for a merchant