Adding a knowledge source and its schedule
Four questions decide a source's whole record: which part of your site this is, what it's about, how often to re-check it, and how far the crawl may go.
Registering a knowledge source is really four decisions, and the form is organized in that order because each one depends on the last: what is this part of your site, what is it about, how often should we look again, and how far may the crawl go.
1. Identity
A display name, a stable slug, which kind of source this is (a crawled municipal site section, an uploaded document, an FAQ set, or a manual entry), its primary language, and its location — a URL for a crawled site, an artefact reference otherwise. The kind is fixed once the source exists: a crawled site and an uploaded document set are read by completely different mechanisms, and every article already filed under this source would be misattributed if that changed later.
2. Description and subject — this is retrieval, not a notes field
The subject you declare here is applied before ranking, as a retrieval partition — a wrong subject doesn't mislabel an answer, it removes the source's articles from consideration entirely, and a resident just gets a fluent answer assembled from whatever was left. Get this field right before worrying about anything else on the form.
You'll also mark the content shape — evergreen, dated, or mixed (evergreen prose with a dated block inside it, which requires naming that block's heading so it can be excluded from what's treated as permanent).
Once a source has been crawled at least once, a "draft description" button offers to write this field from the source's own articles — but it only ever fills the box for you to read and correct, never saves on its own. The titles it drew from are listed underneath specifically so a paragraph that sounds right but was actually written from the wrong dozen pages is visible as wrong.
3. Schedule — stated as a sentence, not a cron box
Pick a cadence in plain terms — every day at a time, every few hours, on certain weekdays, once a month — and the equivalent cron expression is shown underneath for anyone who wants to read or paste it. The timezone sits on the same row as the time on purpose: "02:30" is a different instant in winter and summer, and both DST transitions are handled correctly.
4. Freshness is a different question from schedule
The schedule says how often the source is re-checked. Freshness settings say how long an unverified article may sit before it's flagged, and whether a re-crawl's own changes are auto-approved or held for review. Leave auto-approve off unless you've deliberately decided otherwise — the reviewed path is the safe default, on a new source exactly as on an existing one.
5. How far the crawl may go
Seed URLs, path prefixes to include or exclude, an optional sitemap, and a page budget. The budget has a real floor: it must cover every seed and every enumerated page plus one, or the crawl will stop early on every run and can never notice a page that was actually removed.
The form always saves the whole record, on create and on every edit alike — there's no partial update. That's why the last-run summary and every section above are worth reading before you save an edit for something small like a display name.