DEC-0006 — DXA vertical as the first reference implementation

Context

The model (DEC-0003) and the expert team + scaffold-vertical skill (DEC-0005) exist but had never been exercised against a real domain at scale. We needed a first vertical reference implementation: a conformant OKF template for a recognizable firm type, so a new client in that vertical becomes a fast instantiation rather than a from-scratch build. The chosen vertical is the Digital Experience Agency (DXA) — the user's own domain (DXP/CMS/commerce implementation partner), making it the highest-fidelity place to stress-test the model. See dxa-vertical.

Options considered

1. Defer verticals until an ingestion runtime exists and derive templates bottom-up from real client data.
2. Author a thin, illustrative skeleton (a handful of atoms) just to demonstrate the shape.
3. Author a complete, internally-consistent vertical template — a manifest-governed taxonomy of ~50 conformant stubs spanning every category, authored by knowledge-architect agents in waves (foundation/contract → parallel stub fan-out → QA audit → coordinated cleanup).

Decision

Adopt option 3. Ship verticals/digital-experience-agency/ with 3 foundation files (index carrying the authoritative atom manifest, instantiation manifest, vertical log) and 50 conformant OKF stubs: 15 processes, 9 roles, 3 workflows, 6 systems, 5 policies, 6 artifacts, 5 capabilities, 1 exemplar decision. Every stub is status: draft, confidence: asserted, source: vertical-template. Verified at close: zero dangling links, full manifest conformance, clean security scan.

Add three vertical-specific types, each a genuinely new domain noun that an existing type cannot express without distortion (clearing the model's "minimal but conventional" bar):

• engagement — a bounded, dated SOW instance that contains processes (not a process, which is repeatable; not a workflow).
• client — the organization served; the unit of instantiation and anchor for every grounded engagement (not a role, which is an internal function).
• capability — a sellable service offering that bundles the workflows/processes delivering it (the commercial/portfolio unit, not the how).

Per template discipline, no concrete client or engagement atoms are authored — authoring them would invent client facts. Those types are reserved and made Step 0 of instantiation. capability atoms do ship, because the service portfolio is common vertical structure, not client-specific.

Rationale

• Economic engine. The vertical is the platform's leverage: a new client in this vertical is a fast instantiation (flip stubs from draft/asserted to active/verified), not a build. This is where the learning loop compounds across clients.
• Validation under real load. A complete, edge-connected taxonomy is the only honest test of the model. It worked: the v0 model (Knowledge Model v0) held up — every new domain need was met by reusing core types plus three well-justified extensions; no taxonomy rework was required, only convention tightening (see The produces edge is canonical on the producing process only).
• Template, not client data. Reserving client/engagement keeps the template free of invented facts and makes the dxa-manifest the single, honest contract for what an instantiation must supply. Never fabricate facts; ship structure and edges, ground later.
• Provenance preserved. Uniform draft/asserted/vertical-template provenance makes every stub legibly ungrounded until a real engagement grounds it.

Consequences

• Dossier now has a reusable DXA starting taxonomy and an instantiation checklist (dxa-manifest); the scaffold-vertical skill (DEC-0005) is proven and repeatable for the next vertical.
• The model gains three reserved vertical types; future DXA work and verticals inherit the type set and the conventions surfaced here.
• The first grounded client engagement (not yet run) is the next stress-test — it will exercise client/engagement for real and convert stubs to verified.

Review

Revisit after the first real DXA client is instantiated from this template — confirm the type extensions, the 70%-shared assumption, and the manifest's Step-0 discipline survived contact with live client facts.