Living Company · Interactive 2D Framework PUBLIC REFERENCE DESIGN

The original process map is preserved. Use the controls to trace routes, drill into AI processing, continue into BA / FDD review, or follow manager approval through release.

Original process map with live route trace Click the AI screening area to drill down. During trace, the AI drilldown opens automatically.
Living Company · governed request journey Identity, access scope, decisions and audit remain attached from intake to outcome. 1Person submits ideaPurpose, urgency and sourcedocuments travel together. 2Portal receives itCreates a traceable requestwithout choosing the outcome. 3Record createdOwner, identity and correlationID are preserved. 4 · Governed AI screeningThe gateway resolves the requester, allowed sources and tools.PermissionWho can access?MCP toolsWhat can run?EvidenceWhy this route? DecisionWhich governedroute fits?Human gates stay explicit. Duplicate foundShow the matching capabilityand avoid rebuilding it. Reuse or linkReturn the existing owner,process and evidence. Existing process upgradeRoute to the accountable ownerwith an evidence-backed change. Assess the upgradePrepare scope, risk, valueand the required approval. New process or solutionStart governed discoveryand business analysis. Build a reviewable planRequirements, prototype,owner and release decision. Clarify or stopMissing evidence returns withprecise questions. Revise or closeNo silent rejection and nounapproved commitment. Audit and learning loop Request · identity · permitted sources · tool calls · evidence · human decision · outcome · reusable learning

AI Processing Drilldown

Shows the AI request package, access scope, MCP/gateway architecture, live bot runtime, worked examples, selected runtime and audited return path.

Reference from main intake workflow Scoped retrieval and AI execution

Reference: main intake map

Steps 1-3 already happen outside this drilldown: user submits idea, Living Company front door receives it, and the idea record is created.

This drilldown starts when that record is handed to the AI gateway.

1. Living Company AI gateway/server

Receives the idea record package, validates sensitivity, resolves identity and opens audit trace.

Portal never stores model secrets.

Permission

Whose access?

2. Identity mode

On-behalf-of user or scoped service account.

Service account is limited to required sources.

3. MCP control server

Loads allowed tools, connectors and retrieval policy for this user, task and request type.

Enables or disables tools by config.

4. Knowledge retrieval

SharePoint-maintained process lists are indexed by the on-prem indexer.

Only scoped index results are queried.

5. Scoped context

Combines indexed evidence with request rules, output schema and allowed tools.

Two arrows enter here by design.

6. Runtime router

Selects Copilot agent, OpenAI CLI, Claude CLI or local model.

Based on complexity and data rules.

7. OpenAI CLI subscription

CLI subscription runtime receives scoped context, instructions and response schema.

No broad SharePoint access is passed to the model.

8. Output package

Route, confidence, evidence links, reason and suggested next action.

Security, access maintenance and audit

Maintains group-to-source mapping, local ACL scope, index scope, MCP config, model route and decision reason.

 idea package resolve access MCP policy evidence request rules context route audit back to intake workflow
Phase 1 outcome AI screening completes. Duplicate and clarify/reject return to the user. UPGRADE and NEW are promoted to Business Analyst review.
Promote to BA review
Phase 2 starts Business Analyst receives the promoted package and prepares or reviews the FDD / BRD, cost analysis, SWOT, model matrix and automation design.

Phase 2 - Business Analyst Review and FDD / BRD

After UPGRADE or NEW is promoted to BA review, this stage prepares the FDD/BRD, reviews any Fable 5 draft, evaluates viability, compares human versus AI cost, and recommends the automation approach.

Promoted idea becomes BA review work FDD / BRD analysis package Decision and handoff

1. BA receives promoted package

Only UPGRADE and NEW routes become active BA review work.

Includes idea ID, evidence, promotion reason and audit trace.

Prep route

Fable 5 available?

2A. Fable 5 drafts package

Drafts BRD/FDD, process flow, cost comparison, SWOT and open questions.

2B. BA prepares package

BA creates or validates requirements, cost estimate, SWOT and target operating flow.

BA review

Accept?

3. FDD / BRD package

Business requirement document, functional design document, process flow and assumptions.

Accepted package moves forward.

4. Cost comparison

Human effort versus AI/automation effort, build cost, run cost and expected savings.

5. SWOT / viability

Strengths, weaknesses, opportunities, threats, readiness and stakeholder impact.

6. Model / runtime matrix

Compares local model, Copilot Agent, OpenAI CLI, Claude CLI and human-only handling.

7. Automation design

Target process, proposed automation, required data, integrations, risks and controls.

Viable?

Automate, improve, or hold?

Output

Recommended option and next delivery step.

Audit

Reason logged.

 BA package Fable route reject reason accept BA route accepted package handoff
Phase 2 output Accepted FDD / BRD package, cost analysis, SWOT, model/runtime recommendation, automation design and delivery estimate.
Submit to BA Manager
Phase 3 starts BA Manager reviews the document and estimate, decides whether to reject, refine, approve a PoC, or approve real implementation.

Phase 3 - BA Manager Review, PoC, Associate Director Approval and Release

The BA Manager reviews the BA/Fable 5 package and estimate. The first PoC is built as a Copilot Agent, tested live for one week, refined until bugs and workflow nuances are fixed, then moved to GPU/local production runtime after approval.

BA Manager document and estimate review PoC / build, demo and manager decision Associate Director approval, dashboards and live release

1. BA submits package

FDD/BRD, process flow, cost/SWOT, model matrix, delivery estimate and PoC or implementation recommendation.

BA Manager review

Approve, reject or refine?

2. Delivery route accepted

Manager accepts the estimate for PoC or real implementation and authorizes the next delivery step.

Reject / refine reason

Reason is logged, then BA/Fable 5 revises the document or estimate.

3. Build PoC / implementation

BA or Fable 5 develops from the requirement document, scoped data package and approved delivery estimate.

4. Demo package

BA submits the PoC, demo, presentation, known risks, test evidence and implementation notes.

Manager demo review

Approve?

5. Associate Director app

Swipe left rejects with reason. Swipe right approves for release.

Shows business value, risk, demo link and audit trail.

6. Live implementation

App, product, knowledge board or process automation is implemented live or shared with end users.

Decision audit log

Manager rejection/refinement, demo rejection and Associate Director rejection reasons are permanently logged.

BA Manager dashboard

Scoped to the manager's BA stream: approvals, rejections, refinement reasons, PoC status and demos.

AD dashboard

All approved, rejected and pending items with reasons, owners, benefits, risks and release status.

 Open live knowledge bot runtime

End users / live operation

Production access, knowledge retrieval and AI actions continue through the same scoped Access / Index / MCP architecture.

Click to open live bot runtime.
 Open live knowledge bot runtime

Final step: live bot runtime

The released solution uses the same permission scope, context retrieval, MCP tooling and audit controls defined earlier.

Click to open production runtime view.
 submit approve reject/refine build manager ok swipe left swipe right release controls
Applies across every phase Intake, AI screening, BA/FDD, manager approval, release and live operations all write status, owner, decision reason and evidence.
Control layer
Governance outcome The team can see current status, who owns the next action, whether security/support gates passed, actual cost, feedback and retirement status.

Cross-Cutting Governance and Operations Layer

This layer explains the controls that sit across the full framework: lifecycle status, ownership, PoC success criteria, security gates, monitoring, support, cost tracking, versioning, feedback and decommissioning.

Control register and approval gates Live operations, incidents and feedback Cost, versioning, audit and retirement

1. Any phase event

Submission, promotion, BA review, manager decision, release or live incident creates a governed event.

2. Lifecycle status

Submitted, AI screened, promoted, drafted, manager review, PoC approved, demo approved, AD approved, live or retired.

3. RACI / owner

Requester, AI screening, BA, BA Manager, Associate Director, IT owner, security and support owner are assigned.

4. PoC success criteria

Target metric, acceptance test, user group, cost ceiling and risk threshold must be known before build.

5. Security gate

Data class, permission scope, audit enabled, rollback plan and support owner are checked before go-live.

Feedback loop

User feedback, incidents and knowledge corrections can become a new intake or improvement request.

Monitoring

Tool failures, MCP failures, permission denials, low-confidence AI answers and automation failures are captured.

Support model

Named support owner handles incidents, access issues, model/tool issues and operational changes.

Live operation

The released app, bot, board or automation runs through scoped Access / Index / MCP controls.

Audit evidence

Decision reason, owner, status, evidence links, timestamp, reviewer and correlation ID remain searchable.

Cost and value

Estimated human cost, AI/build/run cost, actual run cost and realized benefit are tracked after release.

Version control

Solution version, prompt/tool config version, MCP config version, release date and change reason are recorded.

Retire decision

Unused, replaced or high-risk solutions can be paused, revised, archived or retired with reason.

Archive package

Final version, decision history, support notes, data sources and decommission reason are preserved.

 log gate go live alert correction cost end