Future AI — build plan (updated 2026-08-07)

The living plan behind the [Future AI — Information Design deck] (../presentations/future-ai-information-design.html). Supersedes the plan section of the 2026-08-04 code audit (/root/FUTURE-AI-AUDIT-2026-08-04.md on slwdev — the audit's findings stand; this file tracks what has moved since).

Shipped since the 8/4 audit

Ford spine v1 — ford.intake_links (slw-ford #1, 2026-08-06): every ingested message now records what it attaches to — publisher firm, companies, sectors. "Everything we've ingested about X" is a query, not a mailbox search. Scope is Ford intake only; it is the working precedent for the fleet-wide mention store, not the store itself.
Newsletter / body-only ingestion (slw-ford #1 + slw-idea-atlas #53): a newsletter emailed to ford@ is reformatted by a model pass into a structured issue and delivered through Magellan's signals door as a newsletter drop. Closes the audit's "no body-only capture" gap; newsletter/article routing moved dow → magellan.
Ford library maturing (slw-idea-atlas #40–#53): market-map detection, podcast episode view, check-a-folder, company-overview rebuild, sidebar regroup.
Propose-and-decide mockup (Michael, this repo): built, retargeted to the App Board Status tab per the 8/4 call, three at-volume risks recorded. Waiting on Shawn's placement call.
Mockups posted (this repo): Ford company key-facts strip, LP Flow deal-header density.

No functional commits since 8/4 in: Webster, Terminal, LP Flow, DealFlow, Podcast, Dewey. The nine audit gaps are otherwise unchanged.

The build, restated (what is still missing)

Mention store — fleet-wide "doc X mentions entity Y" still exists nowhere; Ford's intake_links is the seed pattern to generalize.
Typed tag store — no shared sectors/topics/themes vocabulary; Ford's spine carries sector strings, apps carry flat per-app tags.
Relationships — no investor-in / partner-at / led-round edges anywhere.
Entity creation + cross-app enrichment writer — Webster still never creates entities or writes into apps.
Person extraction from docs/newsletters — still absent in Terminal and Dewey.
Ford adapters for 4 destinations (dow, franklin, lp-flow, dealflow) — mail classifies correctly then sits held. (Body-only capture: done.)
Registry joins — the three app matchers still never consult Webster; Podcast entities remain a private registry.
Card system abstraction — all cards bespoke; no docType→template registry.
Extraction-accuracy evals — no golden set for Terminal extraction.

Phases (unchanged shape, updated status)

Phase	What	Status
1	Shared spine: fleet mention store + typed tags + Webster entity-create & relationship edges — generalizing Ford's `intake_links` pattern	Awaiting green-light (Shawn)
2	Point existing extractors at the spine (Terminal, Podcast, Workbench) + people extraction	blocked on 1
3	Ford adapters ×4 + create-or-enrich writer through staged-import rails	blocked on 1
4	Cards read the spine + badges	blocked on 2–3

Process per phase: brief → workstream build → eval gate → flag → queue review → flip. Biggest risk unchanged: matcher fragmentation — the spine must become the one matcher everyone calls.

Decision queue

Green-light Phase 1 (the spine) — DECIDED 2026-08-07: go. The spine builds as part of the SLW Data Engine — Webster + Dewey unified as one product concept (substrate · registry · spine), services separate. Workstream brief: initiatives/data-engine/BRIEF.md.
Propose-and-decide placement — System Design or portal app hub — and who may approve a direction. Gates Michael's initiative.
sync-team-rules.sh go-ahead (nine repos; carries the mockup-publish rule the other repos lack).
Ford classification — cut the scope table. (The Webster/Dewey half is answered: one Data Engine concept, separate services — see DECISIONS.md 2026-08-07.)
Webster review queue — 1,413 pending links (as of 8/4) at webster.rivent.dev. Ceiling on link quality everywhere.
Michael's access — GitHub read on SKO70SV/slw-ford, slw-webster, slw-dewey; Ford production sign-in; the two cited source-of-truth docs on slwdev.