Field note

One contract, two surfaces

A human page and a machine-readable route should expose the same offer, price, limits, and owner.

The human surface comes first

The public website has to work on its own. A buyer should be able to understand the offer and the seller, inspect proof, send an inquiry, and print the terms without JavaScript. An agent should not be required for any of that.

Semantic HTML is the contract’s first published form, not a degraded fallback.

Machine readability removes guessing

Software should not have to scrape a marketing page when a clean record will do. Markdown, JSON, feeds, and OpenAPI can restate the same facts, but they do not get hidden claims or a lower price.

  • Generate each public representation from the same content record
  • Let public GET routes read, never write
  • Publish OpenAPI only after the documented routes work
  • Skip empty Agent Cards and speculative discovery files

A protocol is earned by a capability

A protocol should answer a real need. MCP makes sense when a callable tool exists. A2A needs a long-running service with task state and cancellation. Machine payment needs a repeatable digital output and a refund path we have actually tested.

If those capabilities do not exist yet, say so. A clear site, a written scope, and an ordinary invoice are more useful than protocol theater.

Inspected sources

This note records the inspected state on 2026-07-10. A linked product or specification may change.