Architect-E¶

Phase 4

Architect-E is planned for Phase 4 (Flywheel) of the Rig. This document describes the target design.

Architect-E closes the feedback loop. It reads production usage data and human feedback, then proposes improvements. It proposes — humans decide.

The Flywheel¶

Production data + user feedback
  → Architect-E analyzes
    → Proposes improvement (GitHub issue draft)
      → Routes to appropriate human:
        - Technical (refactoring, performance) → CTO
        - Product (features, UX) → PM
        - Business (pricing, integrations) → PM
      → Human approves or rejects
        → If approved → enters Conductor queue as ISSUE_APPROVED
          → Dev-E builds it
            → Review-E reviews
              → Deployed to production
                → Monitor-E watches
                  → More production data → Architect-E analyzes → ...

Responsibilities¶

Analyze production metrics (usage patterns, performance, errors)
Read user feedback (Cuti-E conversations, App Store reviews, Discord)
Propose new features, refactors, performance improvements
Estimate complexity and flag technical risks
Answer feasibility questions from PM

Proposal Types¶

Type	Routed To	Examples
Technical	CTO	Refactoring, performance optimization, architecture changes, tech debt
Product	PM	New features, UX improvements, deprecations
Business	PM	Pricing changes, integration opportunities, market positioning

Events Emitted¶

Event	When	Data
`PROPOSAL_CREATED`	New improvement proposed	type, title, rationale, complexity
`PROPOSAL_APPROVED`	Human approved the proposal	approver, modifications
`PROPOSAL_REJECTED`	Human rejected the proposal	reason

Data Sources¶

Source	What It Provides
Grafana/Prometheus	Performance metrics, error rates, latency
Cuti-E SDK	User feedback conversations
App Store reviews	User sentiment, feature requests
GitHub issues	Bug patterns, feature requests
Event Store	Development velocity, failure rates, bottleneck analysis

Guardrails¶

Architect-E proposes only — never creates issues directly
All proposals require human approval before entering the queue
Proposals include complexity estimate and risk assessment
Rate limited: max 5 proposals per week (prevent noise)