Reading your knowledge health dashboard
One screen answers the question every other one leaves scattered: what's broken, what do we have, and is any of it stale.
Six other screens in the Knowledge module each know their own column honestly and none of them answer the question you actually open the console with: how much of this is working right now. The dashboard exists to be that one answer, and it's ordered as a question, not as a data model.
1. Needs attention — first, on purpose
Before any total, the top of the page counts sources failing, sources overdue, sources with section problems, sources producing nothing, articles awaiting approval, and pending revisions. A row of tiles below turns red or amber only when its own count is above zero — a permanently red tile is one nobody looks at twice, so the tone tracks the number rather than sitting fixed.
Every distribution table on this page keeps its empty buckets. "Approved: 0" and "a source that produced no articles" are the two findings most worth seeing, and a table that only lists what it found would hide exactly those.
2. The corpus — what you actually have
Total articles, how many are retrievable, how many topics, how many languages, and how many were never approved. Below that, six breakdowns — by status, by language, by facet, by content structure, by subject, by data class — each a real table so a screen reader announces "Approved, 12" rather than a bare digit.
3. Freshness — a different question from "how often does it crawl"
Stale count, how many are past your own freshness policy, how many were never verified at all, and the oldest verified date on record. This section is downstream of your source schedules (see the guide on adding a knowledge source), not a duplicate of them.
4. Crawl health — one row per source
Last checked, next due, consecutive failures, articles produced, stale count, and the full breakdown of what the last run actually did: fetched, created, revised, unchanged, rejected, gone, and problems. The source name links straight to its editor.
5. Indexing — and what's deliberately missing from it
Total chunks, the share embedded, how many are pending, how many failed. One number this screen does not show is "articles with no chunks" — answering it honestly would mean draining thousands of chunk rows for a single integer, so the dashboard says so in its own footnote rather than guessing.
The awaiting-approval and pending-revisions tiles link straight to the Approvals queue, with the age of the oldest one shown beside the count — so a full queue and an old queue read as two different problems, because they are.