# SCALE — Strategic Capabilities Architecture Linkage Engine

## Purpose

SCALE is a vendor-neutral reference design for an AI assistant that retrieves, links, tests, and explains governed business architecture knowledge. It is intended to make architecture answerable without allowing a language model to invent enterprise facts.

## Core proposition

Traditional enterprise search finds documents. SCALE should traverse typed architecture relationships and return a decision-centered answer with sources, conflicts, gaps, ownership, and confidence.

## Minimum knowledge model

### Core objects

- Capability
- Value stream and stage
- Stakeholder
- Organization unit / actor
- Information concept
- Product / service
- Strategy driver, assessment, objective, and outcome
- Initiative
- Policy / rule
- Metric / measure
- Risk
- Decision
- Application / technology reference
- Source artifact

### Minimum metadata

- Stable identifier
- Name and definition
- Type and level
- Status
- Owner / steward
- Source and source location
- Effective date and review date
- Confidence / authority
- Version and change record
- Security / access classification

### Example relationships

- objective **seeks** outcome
- stakeholder **receives** value proposition
- value stage **delivers** value item
- capability **enables** value stage
- organization **performs or owns** capability
- information **is used by** capability
- initiative **changes** capability
- metric **measures** capability, stage, or outcome
- policy **constrains** capability or stage
- risk **threatens** outcome
- application **enables** capability
- decision **approves or rejects** proposed change
- source **supports** record or relationship

## Six architecture layers

1. **Sources** — governed records and approved artifacts.
2. **Normalization** — canonical identity, vocabulary, levels, provenance, ownership, and time.
3. **Linkage** — typed graph relationships and validation rules.
4. **Retrieval** — question classification, domain selection, graph traversal, and evidence ranking.
5. **Explanation** — concise answer, architecture path, citations, conflicts, gaps, confidence, and owner.
6. **Governance** — human review for changes, access control, versioning, lifecycle, and audit.

## Guardrails

- Evidence before fluency.
- No unsupported enterprise claims.
- Label fact, inference, assumption, recommendation, and unresolved conflict.
- Read-oriented by default; proposed changes enter normal stewardship.
- Respect source permissions and security classification.
- Show stale, missing, duplicate, or contradictory records.
- Provide the smallest useful answer for the decision.
- Keep a human owner accountable for architecture knowledge.

## Minimum viable pilot

Choose one bounded value stream and include:

- the stakeholder and value proposition;
- 5–8 value stages;
- enabling L1/L2 capabilities;
- owners and organization units;
- active initiatives;
- key information concepts;
- measures and known pain points;
- approved source artifacts and decisions.

Test 10–20 recurring questions. Measure answer traceability, completeness, correction rate, time saved, and decisions supported—not response volume.

## Implementation patterns

- Microsoft 365 / SharePoint / Copilot Studio with structured lists, content types, metadata, permissions, and governed actions.
- Architecture platform plus retrieval layer over its API or exported graph.
- Graph database with RAG and a separate document source store.
- Lightweight Markdown/JSON repository for a small practice or pilot.

The platform is secondary. The quality of identity, relationships, ownership, source evidence, and governance controls the quality of the assistant.