Skip to content

AI-assisted authoring and human review

Copy page

This handbook uses AI to assist with planning and drafting, but a page becomes something a reader can rely on only after a maintainer has reviewed the text and its evidence. A paragraph, table, or example does not become a fact merely because the site builds.

  • Planning: split a reader path into topics and mark which claims depend on a specific version.
  • Drafting: write Chinese and English from the same outline so heading depth, warnings, and example intent stay aligned.
  • Automated checking: run validation checks to catch missing translations, broken links, or version numbers that disagree with executed examples.
  • Human review: a maintainer reads the public text and evidence, then corrects the copy, sources, and review date when needed.

Executed image examples report format, dimensions, bands, and the binding and libvips versions that produced them. Four interfaces completing a run means that path succeeded on those versions; encoded bytes and defaults can still differ. The page marks that gap as comparison-pending instead of calling the results byte-for-byte equivalent.

A statement that names govips v2.19.0, libvips 8.17.3, sharp 0.35.5, or pyvips 3.2.0 applies to the named version and review date.

Pages no longer use draft, planned, or approved publication states. Once a page enters the public collection, it is maintained as released content; a substantive rewrite requires another source, example, and lastReviewed check. comparison-pending still means cross-binding pixels or defaults have not yet been compared item by item, not that the page has a publication status.

Public pages focus on verified claims, applicable versions, and evidence boundaries. The project does not ship an online playground, and it does not treat a model’s live output as a publishing path equal to human review.