PRDCT-676: give the Generic Extractor tutorial a reading order, and classify all 39 pages

[Iamfle4ka]

Linear: PRDCT-676 · under the dev-ported Diátaxis rework epic. Form only — no page created, split, moved or deleted, no nav change, no prose rewritten.

The reading order

The Generic Extractor tutorial is a real seven-part progression built on one API, but every part ended mid-thought. Only the index carried a list of what to read; the six parts themselves just stopped. Each now ends with a Next pointer following the nav order:

index → REST → JSON → basic configuration → pagination → jobs → mapping → configuration reference

The last part hands off to the reference instead of dead-ending, and the index keeps its existing ## Next Steps list rather than being given a competing one.

Type markers

All 39 pages now carry the type marker convention already used in src/content/docs/cli/. The tree had zero.

Type	Pages
tutorial	the 7 tutorial pages
explanation	the hub, the parameter map
how-to	running, publishing, incremental, SSH proxy
reference	the remaining 26

The markers deliberately carry no verification date. Nothing here was fact-checked against keboola/generic-extractor — this is a form pass — so each marker says exactly that and points at PRDCT-676. A later accuracy pass can grep -type page and replace the note with a real source and date, instead of inheriting a claim nobody made.

Verification

npm run build clean · audit-phase2 MISSING IMAGES 0, counts unchanged from main · all seven Next targets resolve in dist/.

Anchor count reads 23 in this branch because it is cut from main; those are the pre-existing breakages fixed separately in #1112, not a regression here.

Not in this PR

Sentence-case headings (a large mechanical diff across all 39 pages, better on its own), and everything structural — the cookbook extraction and the auth/pagination folds are gated on an owner decision in PRDCT-674.

🤖 Generated with Claude Code

[linear-code]

PRDCT-676

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
connection-docs	Ready	Preview	Sep 30, 2026 1:55pm UTC

[Iamfle4ka]

Added the sentence-case heading pass (1896028c) — 176 headings across 33 pages. It was listed as "not in this PR" above; it turned out small and provably safe, so it belongs here rather than in a seventh PR.

Anchor safety is verified, not assumed. I built the tree before and after and diffed the id="…" attributes: 476 ids before, 476 after, none added, none removed. Starlight slugifies through github-slugger, which lowercases, so ## Data Type and ## Data type both render data-type. Nothing that links into this tree — in this repo or in the keboola org — can break on it.

Left capitalised: product names (Generic Extractor, Keboola), acronyms (API, URL, JSON, OAuth, HTTP, SSH, AWS, CSV), identifiers that are names rather than prose (Has-More, StrToTime), and the first word after a step number.

Three the first pass got wrong and I corrected: "Configuration with Query parameters" and "API Query authentication" — Query was on my keep-list for the auth method name and leaked into common-noun use — and "encrypted Token example".

Build clean, audit-phase2 MISSING IMAGES 0 and totals unchanged from this branch's baseline.

[Iamfle4ka]

@keboola-pr-reviewer review

[keboola-pr-reviewer-bot]

Verdict: auto_approve (risk 2/5) · profile connection-docs

Auto-approve: a content-bucket form pass (heading sentence-casing, Next pointers, type-marker comments) with no URL breakage.