The campaign pipeline — fragments, honestly listed

The model treats campaigns as first-class containers, and the ruled future includes recognizing them as they happen. Today the machinery is fragments. Checked against the code and the live database on 17 August 2026.

What exists today

Capability	Status	The honest detail
Campaign synthesis	built, manual-only, thin	a script groups a brand's already-analyzed pieces into campaigns with a model. It runs only when a person runs it — nothing schedules or triggers it — and its own header says it is a thin slice, not full campaign intelligence
Campaign evidence machinery	built, barely used	two surfaces exist. The source-evidence state machine holds sixty-four rows, all for one campaign (a coincidence — nothing to do with the sixty-four content types). The deeper intelligence tables hold ten runs across two campaigns, with about 450 claims and just over a thousand evidence references. Real machinery; it has run a handful of times
Campaigns themselves	forty-eight exist	mostly minted in batches by the synthesis run, not recognized live
Detection at analysis time	missing	nothing notices "these five new pieces look like a coordinated push" as content lands

What a build would decide first

• Onboarding versus daily are different problems. A backlog lets you see a whole campaign cluster at once; a live feed may not reveal one until several pieces land or an outside signal arrives — a newsroom press release, a trade article. The recognizer needs both modes.
• Evidence before minting. The ruled shape: a campaign is proposed with its evidence and confirmed, never silently created. The evidence tables exist for exactly this.
• The family ruling feeds it. Campaign membership is the container fact the whole reconciliation leans on — the recognizer is what will actually write it at scale.

Where to look next

• Continuous monitoring — the signals a recognizer would watch.
• The Groupings chapter — what a campaign is, and what it may contain.