Recommended Architecture Evolution

Where the ecosystem should go over 12–24 months, expressed as target architecture (not implementation detail). Produced by the Phase-D Architecture Review. It cleanly separates three horizons: current, recommended (12–24 mo), and long-term vision. All of this is a recommendation — the current architecture is unchanged; direction changes are proposed in ADR 0005.

A · Current architecture (as documented)

Six layers, with reusable capabilities embedded inside venture/app repos and manual portal upkeep.

graph TD
    L1[L1 Shared Skills ✅] --> L2[L2 Intelligence Platform<br/>FIP 🔶 + Lab 🔶]
    L2 --> L3[L3 Intelligence Products<br/>intelproducts ✅]
    L3 --> L4[L4 Applications & Agents<br/>inexisstudios · leadplatform · outreachagent 🔶]
    L4 --> L5[L5 Ventures 🧭]
    L5 --> L6[L6 Customer-facing 🧭]
    note[Reusable assessment engine lives INSIDE inexisstudios;<br/>two opportunity scores; loops unbuilt; portal maintained by hand]
    L4 -.-> note

Characteristics (documented): strong pack-driven reuse; but the assessment engine is trapped in a venture repo, opportunity scoring is duplicated, foundational loops (FIP 1–5, lab core, Hermes) are unbuilt, cross-repo consumption is mostly mock, and the portal is hand-maintained.

B · Recommended architecture (12–24 months)

Introduce a Platform Services layer, extract reusable services, make the pack the single source of truth, and automate the portal. Ventures become thin compositions of services + packs.

graph TD
    L1[L1 · Shared Skills] --> L2[L2 · Intelligence Platform<br/>FIP knowledge only]
    L2 --> L3[L3 · Intelligence Products<br/>packs = single source of truth]
    L3 --> PS[L4 · Platform Services  ⟵ NEW<br/>Assessment · Opportunity/COP · Retrieval · Evaluation]
    PS --> L5[L5 · Ventures & Apps<br/>thin: compose services + a pack]
    L5 --> L6[L6 · Customer-facing]
    ORCH[[portfolio-portal-orchestrator]] -.auto-documents.-> ALL[(the whole ecosystem)]
    OUT[Outcomes] -. one-way Hermes feedback .-> L3

    classDef new fill:#e6ffe6,stroke:#1f7a1f;
    class PS new;

Key moves (each tied to a review finding):

Target dependency direction: consumers depend only on platform-service contracts and pinned packs — never on a venture repo. Feedback (Hermes) flows one-way as new immutable pack versions.

C · Long-term vision (24 months +)

A self-improving, measured, and partly commercialised AI platform.

graph LR
    subgraph Platform["Reusable AI Platform"]
        SK[Shared Skills] --> KI[Knowledge Intelligence]
        KI --> PK[Versioned Packs]
        PK --> SVC[Platform Services<br/>assessment · opportunity · retrieval · evaluation]
        SVC --> AG[Governed agents<br/>OpenClaw as service consumer]
    end
    SVC --> INT[Internal ventures<br/>Inexis Digital · Consulting · Inbound Lanka · City Retreats]
    SVC --> EXT[External customers<br/>Assessment SaaS · licensable packs]
    OUT[Measured outcomes] -->|Hermes closed loop| PK
    ORCH[[Self-documenting portal]] -.-> Platform

Vision characteristics:

• Closed measurement loop — every recommendation is outcome-weighted (Hermes), so quality compounds.
• Thin ventures — launching a venture is composing platform services + a domain pack, not building a stack.
• Governed AI orchestration — agents (incl. OpenClaw) are consumers of well-defined services, under one claims/tone guardrail.
• Self-documenting — the orchestrator keeps the portal true automatically.
• Commercialised surfaces — the Assessment Service, intelligence packs, and skills library become external products (see Strategic Opportunities).

Horizon summary

Boundaries of this recommendation: it changes placement, ownership, and automation — not the proven domain logic. It does not assume any unbuilt loop is built; it recommends the sequence to build them. Adoption is via ADRs, one accepted decision at a time, preserving the evidence-first philosophy.

Related

• Architecture Review · Architecture Risks · Strategic Opportunities
• Proposed ADR 0005 · Portfolio Overview · Principles