The website is a renderer, not the source of truth
A public site is easy to mistake for the record. Pages are what people see, so pages become the place facts are edited. That works until the same organisation also needs feeds, schema, AI discovery files, sitemaps and agent-facing indexes — each rewritten by hand, each free to drift. The durable pattern is the opposite. Entities are validated once. Relationships are resolved once. Then every public representation is compiled from that graph. The website is one renderer among several, not the master copy.
- Published
- 5 Sept 2026
- Reading
- 5 min
- Published by
- Brittek Digital
One source, many channels
Brittek's surface separates the human route from the machine surfaces: `llms.txt`, structured LLM context, site index, feed and schema outputs are produced by a compile step rather than authored as independent pages. Change the governed record, regenerate the channels, refuse invalid data at build time.
That is the same sequence argued in the machine-readable architecture note — access, structure, identity, evidence — applied to publishing itself. Structured data on a page is not enough if the page and the feed disagree.
Why this compounds
- Claims stay inspectable because they share an identity across channels.
- Agents and retrieval systems receive the same facts humans read.
- Removal becomes cheaper: retire an entity once, and every renderer loses it together.
- Quality control can fail the build instead of shipping a contradictory surface.
What this is not
This is not a call to hide craft behind a CMS abstraction. Typography, composition and interaction still matter on the human surface. It is a call to stop treating that surface as the only place truth lives.