The floor is three things: an OpenAPI document that accurately describes the API, the JSON Schema that defines the data moving through it, and interactive documentation generated from those rather than written separately by hand. If an API in your estate does not have all three, that is the first gap to close.
Past the floor, the artifacts each answer a different question. A collection — Postman, Bruno, OpenCollection — captures how the API is actually used, with real requests, real auth, and real sequencing. An APIs.json index tells a machine where all of this lives, along with the terms of service, the support channels, and the pricing. Arazzo describes the multi-step workflows that accomplish an actual outcome. Overlays let you specialize a definition for a particular audience without forking it. Spectral or Vacuum rules encode what your organization considers correct.
Why this matters more than it sounds: artifacts are the single biggest reason the API space has been able to automate as much of itself as it has. Documentation, mocks, tests, SDKs, gateway config, governance checks, catalog entries, and now MCP servers and agent tooling are all generated from them. Every artifact you do not have is work somebody does by hand, inconsistently, and then lets drift.
Treat them as products, not build output. In version control, reviewed, linted, versioned, and published where consumers and machines can find them.
