How AI is applied across API Evangelist and APIs.io. Read my AI disclosure →
API Evangelist API Evangelist
Discovery
Learnings
Guidance
Toolbox
Alignment
API Evangelist LLC

Claimed vs. Shipped

Every standard has two numbers, and they are almost never the same one. There is how many organizations say they use it — the thing you can hear at a conference or read in a job posting. And there is how many actually serve the artifact at a URL you can fetch. This page measures both, with two instruments that share no data, and puts the distance between them on one line. Then it asks the question that decides which standards survive the next decade: what does it cost to even read the specification?

Claimed — the hiring read

Shipped — the artifact read

These are shares of two different populations, measured by two different instruments. They are never averaged, and a standard is never given one blended score. The comparison is the point: the same standard, seen from the demand side and the supply side, in the same week.

The gap

Each line runs from the share of hiring companies that name the standard, to the share of catalog providers that actually serve it. A long line is a standard people talk about more than they publish — or publish far more than they ever mention. Hover any row for the counts behind it.

Claimed — named in what companies hire for Shipped — artifact served, with provider evidence

Who actually wrote the artifact

A file sitting in the catalog proves nothing on its own — this pipeline generated most of them. Only two provenance methods count as the provider's own work: harvested, a byte-identical fetch from the provider, and searched, found published at the provider. Everything this pipeline derived or generated is separated out. So is unknown, which means no authorship was ever recorded — not that the provider failed to publish.

Provider-published Generated or derived here No authorship recorded

The toll booth

A standard an agent cannot read is a standard an agent cannot implement. Of the standards where schema access has been verified, most are open — but a hard core is sold by the page, gated behind membership, or blocked outright. This is not a quality judgement about those bodies. It is a statement about what it costs a new implementer, human or machine, to participate at all.

StandardAccessDomainFormHiring demand

What it costs, by domain

The same question asked of every profiled standard, not just the ones whose schema was fetched: is the specification document itself open, paywalled, or restricted? Domains are ordered by how many profiled standards they hold.

Sources and method

The two instruments. Claimed is the companyCount written into each standard's entry in this catalog from the insights jobs corpus, a full-corpus read with a word-boundary matcher whose needles were confirmed against sampled matched text. Each standard carries its own precision grade; low-precision rows are flagged in the table above. Shipped is read from all/*/provenance.yml, which records who wrote every artifact on disk, and from the Kin Score written to all/*/kin/. Protocols and endpoints — MCP, Agent Skills, A2A — are not files and cannot have file provenance, so they are measured by the Kin Score's own first-party provenance and agent-readiness checks instead, and marked wherever they appear.

What this cannot tell you. A standard measured as unshipped may simply never have had its authorship recorded here — that column is shown rather than folded into the gap, and where it is larger than the published count the row is under-measured, not disproven. The hiring corpus is one quarter, weighted to the Fortune 1000 buy side plus API-first providers, so it reads enterprise demand rather than the whole market. Schema access is only reported for standards where it was actually verified. The licensing profile is a v0.1 rubric and its unassessed rows are shown as their own band rather than hidden.

InputWhat it suppliesRecords

Regenerate with scripts/build-claimed-vs-shipped.py. Generated .