Intelligence flow — Ford, the streams, and the spine

How research becomes coordinated knowledge across the fleet: Ford's three ingestion doors, Terminal and Podcast publishing their durable residue, and the Webster entity spine coordinating companies & VCs across DealFlow and LP Flow, with Dewey as the shared document substrate underneath.

The shape

• Ford ingests through three doors (email to ford@rivent.dev, upload, machine drop). One pipeline whatever the door: archive first → classify → extract per typed playbook. Output is always artifact + datapoints (source page + as-of enforced) + signals (amber dot for uncertain matches).
• Terminal stays the flow. Its relevance-gated publisher sends only durable residue (deal signals, datapoints) into Ford every 15 minutes. The gate checks tracked-status against DealFlow, the portfolio, and Webster — so anything in the DealFlow book automatically passes through.
• Podcast stays standalone on processing, connected on output: episode briefs, key facts, and company mentions publish into Ford, Webster-resolved.
• Webster coordinates the books. Every app keeps its own record; Webster links them to one spine entity — links, never merges. Intel reaches DealFlow / LP Flow records via registry links; nothing writes into a CRM directly.
• Dewey is the substrate — one ingest, one embedding space, one search API for Ford, Quill and Slide Library. Entity mapping is Webster's job, not Dewey's.

The one-page diagram

intelligence-flow.html — also on the portal at rivent.dev/presentations.

Out of scope

• Changing any of the shipped pipelines — this initiative documents and verifies the flow; changes are new initiatives.