CEDX Knowledge · AI
CEDX Knowledge runs the handbook as an inventory: every article has an owner, a trust score and a verify-by date, and the answer it synthesizes on the search screen is built from verified cards only — even when the unverified one ranks higher.
The demo workspace's most-viewed card — the Refund window card, 14K views — is unverified, trust 44, and past its verify-by date. The overview doesn't hide that. It puts the card at the top of the debt queue.
Runs on demo data. Northline is the software's demo workspace, not a customer.
What it is
41 collections holding 990 articles, each row carrying category, verified share, debt and 30-day views — and an owner. Support macros is the largest at 52 articles and 48K monthly views with 64% verified; Compliance runs 91% with zero debt. The difference between the two is management, and the table shows it.
334 articles in inventory, 50% of published verified, 106 past verify-by, 4 published with no owner at all. Trust runs next to the flag: FAQ: data retention sits at 99, verified; the Refund window card at 44, not. You can filter to Debt only and work the list.
Search “refund” and the AI answer is explicit: synthesized from top results, verified cards only — “Based on 1 verified card”, with the Returns & refunds FAQ card cited underneath. The top hit by relevance is the unverified Refund window card. The answer refuses to use it.
Product tour
Not a mockup and not a concept deck. This is what opens at /app/knowledge.
01 — Overview
50% verified against a printed 71–80% target; 106 cards of verify debt, 87 of it high; 67 open answer gaps. The alerts are root-caused — the gap “API webhook retry policy” has demand 210 — and the Do-this-next queue puts retiring the legacy SSO fail-open card first.
02 — Collections
Forty-one collections with articles, verified share, debt and views per row. Field runs 93% verified; QA runs 49% with 7 debt. Both have a named owner — the scoreboard is only useful because somebody's name is on it.
03 — Articles
The inventory view: 334 articles with status, views, trust, verified flag, verify-by date and owner. Archived cards stay in the table — Card: API keys sits Archived at trust 20 — because deletion is not the same as governance.
04 — Search
An index of 361 articles and collections. The answer for “refund” cites the verified FAQ card; the two results rank 92 and 84 by relevance — and the 92 is the unverified card the answer declined to use.
Who runs it
Roles, not references. We have no named customers yet, so nobody in these photographs is quoted, credited or claimed as one.
Works the debt queue in order — 106 cards past verify-by — starting with the policy-critical ones the alerts name first.
106 past verify-by
Owns a collection's verified share the way a lead owns a number. Support macros at 64% has a name next to it, and a 14-card debt count.
41 collections · all owned
Searches “refund”, gets an answer built from the verified card — and flags the unverified twin ranking above it.
answer gaps · 67 openThe shape of it
Every figure below is legible in the captures above. Nothing here is a projection of your estate — it is the state of the demo data.
How it runs
A card starts as a draft in a collection with a category and an owner — 57 drafts sit in the demo workspace now.
Publication sets a verify-by date and a trust score; the W5 SSO verify sprint annotated on the chart is what catching up looks like.
Search synthesizes from verified cards only and cites the one it used — an unverified card can rank higher and still be refused.
Stale cards don't linger: the do-this-next queue's first item is retiring the legacy SSO fail-open card, 154 days past its verify-by date.
One record
An assistant that must cite its sources is only as good as the sources. Verification is what makes a card quotable.
Copilot's system prompt restricts answers to retrieved context — this is the context. Its citation-quality gate exists because verification does.
CEDX DeskSupport macros is the demo's largest collection — 52 articles, 48K views a month — the same macros a desk team works from.
CEDX AgentsAn agent's answer is only as good as what it retrieves. Verified collections are what guardrailed retrieval pulls from.
Limits
Finding this out on the third call is worse for you than reading it here, and worse for us.
Start
Open it right now
Your handbook, verified
Knowledge with the rest of it
Questions
Yes. Every screenshot is a capture of the running build and you can open the same build at /app/knowledge. It runs on demo data — Northline is the demo workspace, not a customer.
A card carries an owner, a verify-by date and a trust score; verified means it is inside its window. 50% of the demo's published articles qualify against a target band of 71–80% printed on the health ring. It is a workflow state, not a legal review.
An answer gap opens — 67 are open in the demo, 80 tracked. The top gap, “API webhook retry policy”, shows demand of 210, and the queued fix is to write the card, not to lower the bar.
A named person per collection — every collection row carries an owner. The four published articles with no owner at all are counted on the Articles header, because unowned is a defect, not a default.
No certification has been issued. What we can evidence about hosting, encryption, tenant isolation and production access is written up on the security page.
Live build, demo data, no card. Then ask what your handbook's verified share would be.