Capability dependencies

Target

This page describes intended behaviour or structure. No implementation is claimed unless a section is explicitly marked Current implementation or Evidence.

Why this page replaces the roadmap

The former roadmap pages (short-term, mid-term, long-term; archived under archive/roadmap/) carried 2026–2029 dates and targets that have passed or cannot be evidenced. By the user's direction of 2026-10-08 the Factory has no calendar roadmap, dates, capacity plan or selected pilot at this stage. This page keeps the one thing a plan needs regardless of dates: the order in which capabilities depend on each other. It makes no commitment about when, by whom, or whether any item is built.

Reading the graph

An arrow means "the target needs the source to be usable". Nodes are capabilities, not services. Status in brackets is the current observed state from the capability matrix in handoffs/2026-10-09-independent-review/estate-current-state.md.

flowchart TB
    subgraph Foundations
        ID["Trusted identity & tenant scope<br/>[foundation exists; Factory binding missing]"]
        PORT["One execution port<br/>[missing; three bypasses today]"]
        REL["Immutable asset release<br/>[partial; file catalogs only]"]
        ENV["Run envelope with run/step/attempt identity<br/>[missing]"]
    end
    subgraph Governed run
        RUN["Durable run, step, attempt, lease<br/>[missing]"]
        APPR["Exact-version approval<br/>[missing]"]
        PROF["Model profile & receipts<br/>[partial; adapters only]"]
        CTX["Scoped context & retrieval<br/>[partial; not isolated]"]
        TOOL["Governed tool executor<br/>[partial; MCP governance opt-in]"]
        USE["Usage reservation & reconciliation<br/>[partial; generic metering]"]
    end
    subgraph Trust
        EVAL["Release-gated evaluation<br/>[partial; evaluator registration]"]
        OBS["Run-attributed observability<br/>[partial]"]
        AUD["Audit evidence<br/>[partial; Audit.NET]"]
    end
    subgraph Writes and delivery
        SBX["Sandboxed workspace<br/>[missing]"]
        DIFF["Reviewable change & delivery receipts<br/>[missing]"]
        GEN["Reproducible generation recipe<br/>[partial; templates real, dispatch missing]"]
    end
    subgraph Expansion
        KG["Knowledge graph & memory lifecycle<br/>[partial process]"]
        MKT["Marketplace & distribution<br/>[missing]"]
        CMP["Compute pools & model serving<br/>[unknown]"]
        A2A["Multi-agent & goal-driven operation<br/>[missing]"]
    end
    ID --> PORT
    ID --> CTX
    REL --> RUN
    ENV --> RUN
    PORT --> RUN
    RUN --> APPR
    RUN --> USE
    PROF --> RUN
    CTX --> RUN
    TOOL --> RUN
    RUN --> OBS
    RUN --> AUD
    REL --> EVAL
    EVAL --> REL
    TOOL --> SBX
    APPR --> DIFF
    SBX --> DIFF
    DIFF --> GEN
    REL --> GEN
    CTX --> KG
    EVAL --> KG
    REL --> MKT
    EVAL --> MKT
    PROF --> CMP
    RUN --> A2A
    APPR --> A2A

Hold "Alt" / "Option" to enable pan & zoom

Dependency table

Capability	Needs first	Why	Source of the dependency
One execution port	Trusted scope	Every entry point must carry the same authenticated tenant/project before it can share policy and accounting	ADR-0015; verification V-A5–V-A7
Durable run	Execution port, immutable release, run envelope	A run must pin what it executes, be reachable only through the port, and be identifiable in every event	ADR-0014; first-slice proposal (2026-10-08 packet)
Exact-version approval	Durable run	An approval binds to a persisted artifact or action hash and must survive restart	Packet FR-06; Cloud Design FUX-05
Model profile & receipts	— (needed by durable run)	A run resolves a profile at acceptance; receipts feed usage	Packet GW-04 / BC-07
Scoped context & retrieval	Trusted scope	Backend filtering before top-k needs the scope on the request	ADR-0013
Governed tool executor	— (needed by durable run)	All tool paths converge before side effects are allowed	Packet GW-05 / PK-10–11
Usage reservation & reconciliation	Durable run	Attribution needs run/attempt identity	Packet BC-05 / PK-17
Release-gated evaluation	Immutable release (and feeds it)	Evaluation evidence attaches to an exact candidate; publication is blocked on failure	Packet BC-22 / PK-16
Observability & audit	Durable run	Correlation needs stable run identity; audit needs decisions to record	Packet NFR-10
Sandboxed workspace	Governed tool executor	Side-effecting tools are introduced only inside an isolated boundary	Original handoff §34–§38
Reviewable change & delivery receipts	Approval, sandbox	A diff is proposed from a sandbox and applied only after an exact-version decision	Packet XL-10
Reproducible generation recipe	Immutable release, reviewable change	Generation output is a reviewable diff from pinned templates and schemas	Canonical saas-factory-generation.md; ADR-0025 (descriptors import into blueprints)
Knowledge graph & memory lifecycle	Scoped retrieval, evaluation	Graph and memory are derived, scoped and reviewed assets	Original handoff §22–§32
Marketplace & distribution	Release, evaluation	Only qualified releases are distributed	Packet BC-27/28
Compute pools & model serving	Model profiles	A profile resolves to an endpoint on a pool	Original handoff §11–§14
Multi-agent & goal-driven operation	Durable run, approval	Agent-to-agent delegation reuses run authority and approval semantics	Original handoff §18, §63–§64

What this page deliberately does not contain

• Dates, quarters, milestones, staffing or capacity.
• A selected first outcome or pilot. The repository-assessment workflow described in the 2026-10-08 packet remains a worked example.
• Any statement that an item has started, is funded or is scheduled.

Where the archived roadmap content went

Archived page	What was kept
archive/roadmap/short-term-plan.md	The foundation ordering (templates stable → foundation agents → orchestration → Studio) is reflected in Foundations → Governed run above
archive/roadmap/mid-term-expansion.md	Domain templates, agent enhancement and vector-search knowledge appear as Reproducible generation and Knowledge & memory
archive/roadmap/long-term-platform-evolution.md	Self-service, marketplace, enterprise features and autonomy appear as Marketplace, Trust and Multi-agent nodes without targets
archive/roadmap/foundation-agents-mvp.md, archive/roadmap/first-saas-mvp.md	Their end-to-end loops are preserved as worked examples in the Software Factory solution pack; their "proves the hypothesis" claims are not evidence

The per-platform extension-roadmap.md pages are kept as dateless "later capabilities" sections inside their plane (user decision D5).