What kind of document the statement was printed in
The genre of the source page decides what can be read off it. A grid settles a crossing point and describes no file; a reference settles a default and describes no plan. One page here per kind of document, listing the entries read from it. As of 2026-09-22.
Six kinds of document supply every value on this site. They are not equally informative, and the differences are not about the honesty of the companies publishing them. A pricing grid exists to make an upgrade look worthwhile, so the crossing point is the one thing on it stated plainly and the contents of a file are structurally absent.
Reading the register by genre separates two groups of company that a product-by-product listing mixes together. One group publishes a tier and no file contents; the other publishes file contents, names a standard, and never mentions a plan. That split follows the genre of the document almost exactly, which is a fact about who each company writes for.
Each page lists only the entries actually read from that kind of document, so the membership of the six lists adds up to the register with no row counted twice. Where one generator is read from two genres, the reason is on both pages and the values are attributed separately.
1The genres
- Plan grids — Nine generators are read from a sales document, which states a crossing point and describes no file.
- Developer references — Two generators whose marking answer is a request field with a published default, or a line under Limitations.
- Model cards — Two generators whose marking, metadata and detection are listed together as safeguards, with no tier anywhere.
- Operator guides — The one genre that publishes a reason for the mark and the pipeline stage it lands at.
- Licence terms — The one document here that says what happens to files generated before an upgrade.
- Research notes — The fullest description of a mark on this site, and the only one naming transformations it should survive.
Ordered from the commonest source to the rarest, which is also roughly the order from least to most detail about the file itself.
Elsewhere in the register: Generators, Fields, Readings, Side by side, Implementers, Questions, Findings, Terms, Learn, Data. Field definitions live on the protocol page.