How these docs stay current

Most product docs rot the same way: someone writes them, the product changes underneath, and nobody circles back. The Concepts, Guides, and Reference categories in this site are built so that can't happen to them — their wording is generated from the application's own source, not typed by hand.

Two stages, one pull request

A script reads three things straight out of the application: the glossary of product terms, the status registries that back every pill and badge you see on screen, and the registry of guided workflows. It evaluates them the same way the app itself does — through the app's own build tooling, not by parsing text — so a renamed field or a deleted screen breaks the generation rather than quietly going stale. That's the extract stage.

The render stage then rewrites clearly marked blocks inside the authored articles from that extracted data — one sentence per line, so a wording change shows up as a small, reviewable diff rather than a reflowed paragraph nobody can review. Everything outside a marked block is normal, hand-written prose; only the blocks a page explicitly opens are ever touched.

Both stages run locally as part of writing the change, so the generated text lands in the very same pull request as the code it describes — never a follow-up someone forgets to send.

What keeps it honest after that

A pull request that touches the glossary, a status registry, the guide registry, or the generator itself runs a check that regenerates everything and fails if the result doesn't match what's committed. There's no path for the product to change and the docs to fall behind: the two are either coherent, or the build is red.

The rest of the site — this page included, and the Getting Started walkthrough — is ordinary hand-written prose. The team is held to keeping it accurate the old-fashioned way; only the vocabulary and status pages carry the stronger, generated guarantee.

Tell us if it's wrong

Every article on this site carries a Was this helpful? control at the bottom. It doesn't require an account, and a "No" tells the team an article needs a second look even when nothing here caught it first.