How this site is built as evidence.

Two claims are byte-frozen.

The site holds two sealed claims — the definitions of the Evidence Economy and of the Operator Gap. A sealed claim appears exactly once on its canonical page, wrapped in an HTML pattern that identifies it by data-claim-id, declares its class, and marks it sacred and non-editable. The same claim is published in machine-readable form at /api/sealed-claims.json and referenced from /ai-context.jsonld.

A build-time check verifies that the inner text of the sealed block is byte-identical to the envelope. If the check fails, the site is not deployed.

The HTML pattern is not invented for this site. It is the same pattern used by NODE-18 for its canonical Geometry Intelligence definition,[S1-NODE18-CANONICAL] reproduced verbatim so a consumer that already reads NODE-18 can read NODE-20 without a special case.

Six routes for machines.

Alongside the HTML pages, the site publishes six machine surfaces: /robots.txt, /sitemap.xml, /llms.txt, /ai-context.jsonld, /api/node.json, and /api/sealed-claims.json. Each is generated from the same data files that render the HTML — there is no manual copy of a claim text or of a source URL anywhere else in the codebase.[S2-JSONLD][S2-SDO]

Notably absent: /mcp.json. NODE-20 does not publish an MCP endpoint because it does not host an MCP server. Publishing an mcp.json for a non-existent endpoint would be a decorative surface, not evidence.

Publisher, author, dates.

Every page carries provenance in three places: the visible footer (with the publisher's Wikidata QID and the originator's Wikidata QID),[S1-WD-KTS][S1-WD-TJ] the JSON-LD graph embedded in the page, and the KTS-federation meta tags in the document head. The distinction between publisher, originator and evidence anchor follows the entity/activity/agent shape of the W3C PROV data model at a conceptual level, though PROV records themselves are not published.[S2-PROV]

Every artefact says when it was frozen.

A consumer that reads a KTS node three months later should be able to determine which snapshot of the federation it is looking at. Two markers do that work: built_at on the node record, and topology_version_observed — the version of the NODE-1 topology at the moment this node was built. Both appear in the footer of every page and in the machine surfaces. NODE-20 records the observed topology version as 2026-05-26-v3.2.1-fabric.

Explicit boundaries.

• Not a benchmark. Nothing on the site ranks any organisation's evidence quality.
• Not an accreditation programme. There is no seal, no certificate, no register of accredited members.
• Not proprietary market research. Where the site describes AI-mediated discovery, it cites the public documentation that describes it.[S2-SEO-AI]
• Not a case-study repository. Client outcomes are not asserted here.

Sources