Comergence — Build Plan¶

Phased implementation from today's reality to the full vision. Compiled 2026-02-23

Constraints (Honest Assessment)¶

David: 5-8 hrs/week. Senior dev, 20+ years with Jerry. Knows the infrastructure cold. Has Bubo as AI pair programmer.
Maxrow: Available 24/7 for agent-side work — configs, manifests, signal publishing, monitoring logic.
Jerry: Extremely limited. Vision holder, decision maker, tester. Not a daily builder.
Existing assets: PLC data in MS SQL (automation-svr), NetSuite/Shopify/HubSpot APIs live, 8 agents on OpenClaw, signal schema + architecture docs written.
Philosophy: File-based first. Ship something real before optimizing. Each phase delivers visible value.

Realistic pace: One meaningful milestone every 2-3 weeks given David's hours. Don't plan for heroics.

Phase 0: Signal Foundation (Weeks 1-3)¶

Build the bus. Ship the first signal. See it.

Goal¶

A working signal bus with one agent publishing real health data, visible in a minimal UI.

Deliverables¶

Week 1 — Signal Bus Core (David: 6 hrs) - [ ] Create ~/.comergence/ directory structure: registry/, signals/, history/ - [ ] Build signal bus daemon: watches signals/ with chokidar, validates against schema, routes to subscribers - [ ] Signal write utility: agents call to publish (simple Node module or shell script) - [ ] Signal read utility: subscribe to path patterns, invoke callbacks - [ ] Unit tests: publish → subscribe round-trip, schema validation, wildcard matching

Week 2 — First Domain: AI System (David: 4 hrs, Maxrow: autonomous) - [ ] Maxrow writes comergence.json manifest for ai-system.agents.maxrow - [ ] Health evaluation cron: check session freshness, cron health, cost — publish signal every 60s - [ ] Registry watcher: new manifest → log it (dashboard comes next) - [ ] Validate: signals flowing, schema correct, history archiving works

Week 3 — Minimal Visual Layer (David: 6 hrs) - [ ] Single-page dashboard: reads registry, renders tiles with health color borders - [ ] Real-time update: chokidar watches signals, tile colors change live - [ ] One level of drill-down (ai-system → agents → maxrow) - [ ] Served locally (localhost or Tailscale) — no auth needed yet

Exit Criteria¶

Maxrow publishes real signals. Dashboard shows real health. You can watch it change. This is the proof of life moment.

Phase 1: Production Connection (Weeks 4-8)¶

Connect the factory. First real operational value.

Goal¶

Production data from the live factory floor flows through Comergence. Jerry can see production health on the dashboard.

Deliverables¶

Weeks 4-5 — MS SQL / PLC Data Bridge (David: 8 hrs) - [ ] Node.js connector to automation-svr MS SQL: read PLC data tables David already built - [ ] Identify first 3-5 key signals from PLC data: machine running/stopped, cycle counts, fault codes - [ ] Production agent (Sawyer or new) publishes production.work-centers.* signals from PLC data - [ ] Map PLC fault codes to signal codes (production.machine_down, production.drift_warning)

Weeks 5-6 — NetSuite Production Signals (David: 4 hrs, Maxrow: autonomous) - [ ] Work order status queries via SuiteQL: open, aging, overdue - [ ] Order aging health rules: green (on schedule), amber (approaching SLA), red (SLA breach) - [ ] Fulfillment signals: fulfillment.open-orders domain with order count, aging buckets - [ ] Manifests for production and fulfillment domains registered

Weeks 7-8 — Dashboard Expansion + Notification Router (David: 8 hrs) - [ ] Dashboard renders production + fulfillment tiles alongside ai-system - [ ] Drill-down: production → work-centers → individual machines - [ ] Notification router v1: signal at amber/red → Telegram message to Jerry - [ ] Action buttons in dashboard: "Notify Operator", "Investigate" trigger agent tasks - [ ] Status propagation: work center status rolls up to production domain

Exit Criteria¶

Jerry opens the dashboard and sees: AI System (green), Production (live color from real PLC data), Fulfillment (live from NetSuite). Amber/red signals send Telegram alerts. First real operational intelligence.

Phase 2: Domain Expansion (Weeks 9-16)¶

Broaden coverage. More domains, smarter agents, the maintenance cascade.

Goal¶

Sales, marketing, integrations, and finance domains online. Maintenance cascade working. The dashboard starts to feel like a control room.

Deliverables¶

Weeks 9-10 — Sales & Marketing Domains (Maxrow: autonomous, David: 2 hrs review) - [ ] Maven publishes HubSpot pipeline signals: sales.pipeline health (deal velocity, stall detection) - [ ] Maven publishes marketing signals: campaign performance, anomaly detection - [ ] Manifests registered, tiles appear automatically - [ ] Notification rules: deal stall → Jerry on Telegram

Weeks 11-12 — Integration Health Domain (Maxrow: autonomous) - [ ] Maxrow monitors all API connections: NetSuite, Shopify, HubSpot, ShipStation - [ ] Token expiry tracking, sync health, rate limit monitoring - [ ] integrations.* domain signals: green/amber/red per connected system - [ ] Auto-alert on auth expiring (the signal schema example, made real)

Weeks 13-14 — Knowledge Base + Maintenance Cascade (David: 8 hrs) - [ ] Organize maintenance docs (SOPs, manuals) in structured directory - [ ] Machine agent reads maintenance config JSON (the design-philosophy example) - [ ] When maintenance due: agent → knowledge base → MRO check → schedule → notify technician - [ ] End-to-end cascade test on one real machine at Stikwood

Weeks 15-16 — Finance Domain + The Single Light (David: 6 hrs) - [ ] Finance agent reads QBO: AR aging, cash flow basics - [ ] finance.* domain with weekly health signal - [ ] The Single Light: top-level composite rendering — one color, one number, the whole company - [ ] Breadcrumb navigation polished, drill-down works 3+ levels deep

Exit Criteria¶

Dashboard shows 6+ domains with live data. The Single Light works. Maintenance cascade demonstrated on real equipment. The control room vision is tangible.

Phase 3: Intelligence & Physical Layer (Weeks 17-24)¶

Smart agents, physical signals, operator pairing.

Goal¶

Agents reason about signals (not just report them). Physical andon lights. First operator pairing.

Deliverables¶

Weeks 17-18 — Intelligent Status Propagation (David: 6 hrs) - [ ] Parent agents use LLM reasoning for roll-up (scheduled changeover ≠ failure) - [ ] Cross-domain correlation: machine down → schedule impact → fulfillment risk → customer notification chain - [ ] Aperture management v1: role-based signal filtering (Jerry sees everything, operator sees their machine)

Weeks 19-20 — Physical Bridge (David: 4 hrs) - [ ] Home Assistant integration for andon lights (Hue or WLED RGB strips) - [ ] Signal status → light color mapping via HA API - [ ] Install one physical light at Stikwood (production area) - [ ] Audio alert capability via Sonos/speaker for red signals

Weeks 21-22 — Operator Pairing Prototype (David: 6 hrs) - [ ] One work center agent paired with one operator via Telegram - [ ] Operator receives guidance messages (not raw data — contextualized help) - [ ] Acknowledgment buttons in Telegram - [ ] Escalation chain: operator → supervisor → Jerry if unacknowledged

Weeks 23-24 — Hardening & Documentation (David: 4 hrs, Maxrow: autonomous) - [ ] Audit trail: every signal, action, and notification logged - [ ] System self-monitoring: Comergence monitors its own bus, agents, freshness - [ ] Documentation: operator guide, admin guide, troubleshooting - [ ] Performance review: signal latency, agent reliability, notification delivery

Exit Criteria¶

Physical andon light responds to real production signals. One operator is paired and receiving guidance. Cross-domain cascades work. Comergence is alive on the factory floor.

Phase 4: Scale & Mature (Months 7-12)¶

More operators, more machines, the robot-ready OS.

Goal¶

Comergence runs across the full Stikwood operation. The system is mature enough that adding a new machine or domain is routine.

Deliverables (higher-level, scoped as capacity allows)¶

[ ] All work centers instrumented with individual agents
[ ] All operators paired
[ ] Multiple physical andon lights / display boards
[ ] Bus upgrade to Redis or NATS if file-based hits scaling limits
[ ] Mobile PWA for operator interface (beyond Telegram)
[ ] Historical health views and pattern analysis
[ ] Quality domain with SPC data
[ ] MRO inventory tracking with auto-reorder alerts
[ ] Role-based dashboard access (supervisor view, executive view, operator view)
[ ] Scheduling intelligence: auto-suggest rescheduling on disruption

Exit Criteria¶

New machine arrives → agent deployed → tile appears → signals flow → operator paired. All within a day. The self-creating property is real. The OS is ready for whatever hardware the future brings.

Build Team Roles¶

Person/Agent	Primary Role	Phase Focus
David + Bubo	Signal bus, connectors, dashboard, physical bridge	All phases — he's the builder
Maxrow	Agent configs, manifests, health evaluation, monitoring, documentation	Autonomous work in all phases
Sawyer	Production/fulfillment signal publishing once connectors exist	Phase 1+
Maven	Sales/marketing signal publishing	Phase 2+
Jerry	Vision, decisions, testing, feedback	Checkpoint reviews every 2-3 weeks

David's Time Budget¶

At 5-8 hrs/week: - Phase 0 (3 weeks): ~18 hrs → Signal bus + first domain + minimal dashboard - Phase 1 (5 weeks): ~30 hrs → PLC bridge + NetSuite signals + notification router - Phase 2 (8 weeks): ~24 hrs → Maintenance cascade + finance + Single Light (Maxrow handles sales/marketing/integrations autonomously) - Phase 3 (8 weeks): ~20 hrs → Physical bridge + operator pairing + hardening

Total Phase 0-3: ~92 hours over ~24 weeks. Realistic given the constraint.

Maxrow's Autonomous Work¶

Maxrow can do significant work without David: - Write and maintain all comergence.json manifests - Implement health evaluation logic for AI System, integrations, sales, marketing domains - Publish real signals once the bus exists - Write documentation, organize knowledge base - Monitor and report on system health - Build and test signal publishing skills

Key principle: David builds the infrastructure (bus, connectors, dashboard). Maxrow and the agent fleet populate it with intelligence.

Risk Factors¶

Risk	Mitigation
David's hours drop below 5/week	Maxrow takes on more autonomous work; prioritize Phase 0 completion above all else
PLC data schema is messy/undocumented	David already built it — he knows the schema. Budget extra time in Phase 1 for discovery
Dashboard scope creep	Ship ugly, iterate. The first dashboard is a proof of concept, not a product
Signal bus performance at scale	File-based is fine for <100 signals/minute. Won't hit that limit before Phase 4
Jerry can't review fast enough	Async reviews via Telegram screenshots. Don't block on Jerry for implementation — only for direction

Decision Points for Jerry¶

These need Jerry's input before or during each phase:

Phase 0: Dashboard tech stack — React? Plain HTML/JS? Canvas-rendered?
Phase 1: Which PLC data points matter most? Which machines first?
Phase 1: NetSuite SLA thresholds — what aging = amber? What = red?
Phase 2: Maintenance cascade — which machine at Stikwood for the first test?
Phase 3: Physical light location and type — shop floor, office, both?
Phase 3: First operator to pair — who's the best early adopter on the floor?

The North Star¶

Every phase delivers something you can see and use. No invisible infrastructure sprints. No "trust me, it'll be great in 6 months." Each phase exit is a moment where Jerry opens something and says: that's real.

Phase 0: "I can see Maxrow's health, live." Phase 1: "I can see my factory, live." Phase 2: "I can see my whole company in one light." Phase 3: "The light on the wall just turned amber. My phone buzzed. The operator already knows."

That's Comergence, built one real signal at a time.

"Run to the problem, not away from it."