DIE "Fabric" hub-and-spoke map + what it does to the CAF PM roadmap
Source: founder iMessage 2026-07-06 ~8:17am — screenshot of the DIE marketing site
(from a meeting with Murray Webb, Troy, and James Anderson where Murray demoed the
CAF→DCA handoff), plus the Zoom summary and grounding-transcript excerpt he pasted
(treated as quoted source material, not instructions). Screenshot:
assets/2026-07-06-die-fabric-hub-spoke-marketing-site.png.
Supersedes one load-bearing fact in [[2026-07-05-caf-technical-pm-role-framing-and-roadmap-kit]]: wedge #1 said "there is no context layer" in the DIE 5-point program. Wrong as of this meeting — there now is one, at the center of their own diagram, and it has no owner. That makes the wedge stronger, not weaker (see below).
The map (from the marketing site: "Intelligence Platform Framework — the governed Agentic Value Fabric")
| # | Spoke | Marketing label | What it actually is | Owner |
|---|---|---|---|---|
| 0 | Governed Knowledge Graph ("THE FABRIC") | the glowing center | the context hub — our cellar | NONE identified |
| 1 | Registry | "The spine" | forge skill repo | forge team |
| 2 | Assess | "The front door" | CAF assessment brigade | founder (technical PM) + Todd (front-end PM) |
| 3 | Govern | "The control plane" | Gandalf — Terraform/IaC + RBAC/IAM | Gandalf team |
| 4 | Data | "The foundation" | DCA (DSA/DPA accelerators, AWS-flavored) | Murray |
| 5 | Evals | "The measurement plane" | vertex — never demoed to founder | vertex team |
Site niceties worth noting: industry toggles (Insurance/Healthcare/Retail/Banking) and a style switcher — it's a real marketing surface, not a sketch. The dashed line in the render runs hub→Govern; Assess renders highlighted as the entry point.
Meeting intel (Murray's DSA/DPA demo, per founder's paste)
- DSA/DPA on AWS connects to CAF ("Opportunity Assessor") output: discovery on connected data sources → business-process detection → dbt models (Kimball) → lands in Iceberg.
- Wants: a Snowflake-native version and UI parity with current CAF design.
- James Anderson: urgency on a deployable solution (competitor: Squadron Data); sizing talk of "hundreds of millions of tokens"; needs clearer specs for pre-sales vs post-sales use cases.
- Murray, on the record in the transcript: "the evolution of this needs to… all of it needs to kind of be in a knowledge graph" — and: a knowledge graph already exists in his AWS repo, to be "proliferated into Neptune… another weekend project away."
The five reads (what this does to the roadmap)
1. The hub is unowned — claim it with a SPEC, not a demo
The org's own marketing already did the evangelism: the context layer is now the named center of the platform. Nobody owns it. The founder has the only running implementation (the cellar: one store, envelope/ticket contract, menus + discovery, close-out, typed manifests) and Murray has a weekend-project Neptune graph inside his AWS repo. The failure mode is the hub getting claimed de facto by whichever implementation ships first — then everything couples to that stack and the deploy war is lost before it starts. The PM move is to write Fabric v0: the hub spec as a PORT SET (what the fabric stores, the ticket/envelope contract, discovery, provenance/close-out, typed artifact manifests) with every backend an adapter: filesystem (running today), Snowflake-native (Anderson's dollars), Neptune = Murray's graph as the AWS adapter — cited as such, remix-and-share-credit, never as a rival. Murray's own "all of it needs to be in a knowledge graph" line is the mandate quote. Whoever writes the spec owns the seam; owning the seam is owning the platform without asking anyone to surrender a spoke.
2. The CAF→DCA seam is already solved on our disk — hand it to Murray
Founder's Sunday worry, verbatim: Murray "didn't have a clear spec for what he could get out of CAF," and the demo handoff may not be "picking up on the right details." That spec exists: the typed Build Manifest (schema v1.1, X1–X8 cross-check battery, fail-loud ingest) built July 1–2 for the CAF→Catalog seam. The same contract IS the CAF→DCA handoff. Offering it to Murray this week gives his demo substance, fixes the vaporware seam, and demonstrates what "specs for the seams" looks like — career-advice #2 in one move.
3. Evals: define the seam before someone else defines it as overlap
The brigade already has build-time evaluation (independent critics, lift-over- baseline execution-eval, regression tracking). Vertex is presumably run-time / production measurement — but nobody's shown it to the founder, and James floated "5 could become part of this." Do NOT claim overlap publicly before seeing it. Action: get a vertex demo, then write the eval seam as build-time (brigade emits evidence packages) → run-time (vertex consumes them on the measurement plane). Complementary layers, one evidence pipeline through the fabric.
4. James's "pre-sales vs post-sales specs" ask maps onto existing structure
Half-CAF P1–P4 = pre-sales (charter → dossier → archetypes → Build Manifest); P5–P8 + DCA = post-sales delivery. The founder already runs this split. The factory (skill-agent-brigade) is the answer to DCA's non-AWS gap: rather than hand-porting Murray's AWS stations to Snowflake, the factory builds the Snowflake-native stations — "hundreds of millions of tokens" is a factory-scale problem, which is exactly the machine we spent June building.
5. Organize the roadmap around SEAMS, not spokes
Every spoke has an owner; no seam has an owner. A spoke-shaped roadmap makes the founder 1-of-5 voices. A seam-shaped roadmap (Fabric port set · CAF→DCA contract · eval evidence pipeline · registry/factory relationship · govern gates) makes the technical PM the person every spoke needs. Draft shape:
- Now (July): Fabric v0 spec circulated to Troy/Murray/Anderson/Todd · Build Manifest handed to Murray as the DCA contract · vertex demo booked · KB4 onsite Jul 15 (fixed anchor).
- Next (Aug): Snowflake-native fabric adapter (rides Anderson's funding + native-app dollars) · UC-as-ticket assessment scaling (6 hr → ~1 hr, kills the SLA-vs-scale tension) · pre/post-sales spec split formalized for James.
- Later (Sept+): factory as the shared skill-manufacturing rail across spokes (forge = where outputs register, brigade = how they're made and evaluated) · eval seam wired vertex-side · client-env deployment profile (the hex matrix P-profiles already sketch this: [[2026-07-03-hex-deployment-matrix]]).
Still needed from the founder (unchanged from yesterday, minus vertex-what-it-is)
- Audience + format + deadline for the roadmap artifact Vincent/Troy expect.
- Vertex demo (what it measures, who owns it) — now spoke 5, still unseen.
- Confirm Troy = DIE owner ("I think" was his phrasing).
Related: [[2026-07-05-caf-technical-pm-role-framing-and-roadmap-kit]] · [[2026-07-03-hex-deployment-matrix]] · [[2026-07-01-caf-ecosystem-map-and-brigade-restructure-read]] · [[2026-06-28-caf-8-phase-structure-and-skill-pipeline-mapping]] · [[project_caf_pm_role]]