Sovereign Orchestrator Pro V6.0 is a 173-file, ~38,190-line TypeScript system that autonomously generates, quality-gates, and publishes marketing content across WordPress, Shopify, X.com, and Reddit. It runs a tri-agent War Room for content consensus, a VERITAS anti-hallucination layer, and hybrid Ollama/Gemini model routing — all operated from a single React dashboard.
- 173 source files and ~38,190 lines of code — verified by programmatic file inventory across the
src/directory. - 18 specialized execution threads with a performance-ranked thread selector that picks based on historical daypart performance.
- 5-layer content gate stack: cadence (B1) → substance (B3) → novelty → VERITAS → brand policy. No content publishes without clearing all five.
- 9+ external integrations (Shopify GraphQL, WordPress REST, X.com, Reddit, Klaviyo, Resend, IMAP, Gemini, Ollama) each with Iron Dome rate-limit protection.
- 8 of 11 subsystems healthy (green), 2 at-risk (yellow), 1 known-issue (amber) as of the June 2026 Master Audit.
- Equivalent market value $100k+ — actual build compressed to ~1–1.5 months via AI-orchestrated development.
- Honest trade-offs documented: stealth DOM brittleness on X/Reddit, unbounded Hippocampus memory, partial test coverage on Playwright workflows.
Sovereign Orchestrator Pro V6.0 demonstrates that a solo builder with AI-assisted development can ship a production-grade, multi-platform marketing automation system that would cost $89k–$350k+ at traditional agency or senior-engineer rates. The system is running — not a prototype — with documented trade-offs and a prioritized remediation roadmap.
1 · What Is an Autonomous Marketing Orchestrator?
Most marketing automation tools handle scheduling and template-based posting. SOP V6.0 goes further: it generates the content, argues over it in a multi-agent War Room, verifies every claim against brand policy (VERITAS), and publishes to the appropriate platform with rate-limit protection (Iron Dome) — autonomously.
The system also manages product reviews (IMAP intake → sentiment extraction → Shopify metaobject sync → discount-reward email), sends multi-channel newsletters (Klaviyo, WordPress subscribers-only, Resend), and learns from its own performance through a vector memory system called the Hippocampus.
The entire operation is controlled from a React “Nerve Center” dashboard with a one-button Start/Stop state machine and real-time telemetry.
2 · System Architecture
At the core is the Orchestrator Core, controlled from the React Nerve Center. It spawns up to 18 specialized execution threads partitioned into AM and PM blocks by a daypart guard. A performance-ranked thread selector dynamically picks threads based on historical performance data.
The AGI Swarm Configuration Picker routes tasks to either Ollama (local, zero marginal cost) or Gemini Cloud (cutting-edge reasoning) based on availability. Complex outputs pass through a Tri-Agent War Room where distinct agent personas critique, argue, and reach consensus.
Every approved proposal passes through VERITAS — the system’s anti-hallucination constraint layer — before publishing to any platform. The Iron Dome resilience layer intercepts cascading API failures.
3 · The Content-Quality Gate Stack
Ensures content is not over-published. Checks against the current daypart and recent posting limits.
Measures depth and relevance. Rejects content scored as “filler” using a structural rubric with a minimum quality floor.
Compares the proposal against recent outputs stored in Hippocampus memory. Duplicate or near-duplicate angles are rejected.
Strict anti-hallucination layer. Verifies all factual claims. No invented numbers, no fabricated certifications, no unverified product specs.
Enforces: explicit ban on OEKO-TEX (only GOTS, GRS, OCS, PETA-Approved Vegan, Fair Trade allowed); scarcity ban (no fake urgency); live-only links (all internal URLs must return HTTP 200 before use).
4 · Hybrid Model Routing
The AGI Swarm Configuration Picker (src/agents/agentModels.ts) makes routing decisions based on:
- Ollama (local): High-bandwidth continuous learning, sensitive reasoning, and tasks where privacy matters. Uses
qwen2.5-coder:14bon an RTX 3060 12 GB with CUDA 8.6. - Gemini Cloud: Complex multimodal tasks, image generation (
geminiImage.ts), and when local inference is unavailable. - Cloud-bypass mode: When Ollama is available and the operator enables bypass, the system runs entirely locally with zero API cost.
This hybrid approach means the system can operate autonomously even during API outages or quota limits — a critical design constraint for a 24/7 marketing engine.
5 · The 4 Platform Integrations
Full REST client (wordpress.ts). Subscriber-only posts with auto-private after newsletter send. Jetpack stats integration for telemetry.
Admin GraphQL via shopify.ts. Metaobject sync (shopify-metaobjects.ts). Product, collection, and trust threads. Robust pagination handling.
Headless stealth via xcom-stealth.ts. Engagement builder, content engine, news desk, article desk, and growth engine threads. Rate-limited by Iron Dome.
Stealth browser via reddit-stealth.ts. Discovery, publisher, growth engine, and karma builder threads. Shadowban detection (redditVisibility.ts).
6 · Reviews → Reward Loop
An IMAP client (imapflow.ts) monitors a dedicated inbox for incoming review notification emails.
The review processor (review-processor.ts) extracts sentiment, rating, images, and reviewer identity. Sanitization via reviewSanitize.ts.
Approved reviews sync to a Shopify metaobject for storefront display. Buyer verification checks order history against email.
A Resend email (reviewRewardEmail.ts) delivers a live discount code (thank-you-review) to the reviewer. End-to-end, no human action needed.
8 · Intelligence & Memory
The system does not just execute; it learns. Two subsystems drive this:
- Intelligence Engine (
intelligenceEngine.ts): Analyzes performance data during API downtime or idle periods. Generates structured insights (intelligenceInsights.ts) about which content types, platforms, and posting times perform best. - Hippocampus Vector Memory (
hippocampus.ts): Embeds past successes, rules, and interaction data into a localized vector store underdata/brain/. Agents query this memory to maintain context across days and thread runs, behaving like a stateful intelligence rather than a stateless API call. - Agent Scorecard (
agentScorecard.ts): Periodically generates performance evaluations of each agent persona in the AGI swarm. - Reflection (
performanceBriefing.ts): Stores and retrieves reflection data. Thread-specific reflections are stripped at publish boundaries to prevent leakage.
9 · The Nerve Center Dashboard
The Nerve Center (Dashboard.tsx, ControlPanel.tsx) is not a consumer-facing app — it is an operations UI built for a single operator. Key panels:
- Start/Stop state machine: One button toggles the entire orchestrator. Cloud-bypass mode can be enabled from the same control.
- Cost Monitor (
CostMonitor.tsx): Real-time tracking of Gemini API token spend vs. Ollama local inference savings. - Content Ledger (
ContentLedgerDashboard.tsx): Shows all published content across platforms with status and telemetry. - Shopify Dashboard (
ShopifyDashboard.tsx): Product, collection, and metaobject sync status with Recharts visualizations. - Sniper Dashboard (
SniperDashboard.tsx): X.com reply-sniper targeting with scored tweet queue. - Token Equalizer (
TokenEqualizer.tsx): Visual representation of token allocation across agent personas.
10 · What’s Running Right Now
| Subsystem | Status | Evidence |
|---|---|---|
| Content Pipeline | Healthy | War Room successfully negotiates and constrains drafts. |
| WordPress | Healthy | REST API stable. Subscriber-only + auto-private enforced. |
| Shopify | Healthy | Admin GraphQL pagination robust. Metaobjects syncing. |
| X.com | At-risk | Stealth Playwright is brittle to UI layout updates. |
| At-risk | Same DOM brittleness. Needs selector-healing fallback. | |
| Reviews Intake | Healthy | IMAP parsing stable. Resend reward active. |
| Newsletters | Healthy | Klaviyo null-sender bug resolved via audience-attach gate. |
| Promotions | Healthy | Discount resolution running without rate limits. |
| Scheduler & Resilience | Healthy | Cloud-bypass / Ollama fallback functional. Iron Dome active. |
| Dashboard | Healthy | UI reflects telemetry. Start/Stop responsive. |
| Intelligence / Memory | Known issue | Hippocampus vector store accumulates indefinitely. Lacks TTL. |
Source: Master Audit, June 2026. Audited by Google Antigravity Agent across 173 functional source files.
11 · What’s Possible
SOP was built for a specific brand (Design Delight Studio), but the architecture is general. The pattern decomposes into:
- Orchestrator + Thread Selector: Any system that needs to schedule and prioritize multiple concurrent workflows.
- War Room + VERITAS: Any domain where AI-generated content needs multi-perspective review and constraint enforcement (legal, medical, financial, brand).
- Iron Dome: Any application that consumes multiple external APIs and needs graceful degradation under rate limits or outages.
- Hippocampus: Any agent system that needs to persist context across sessions without re-prompting from scratch.
The constraint is not the code — it is the architecture decisions. The 5-layer gate stack, the War Room consensus mechanism, and the hybrid model routing are the defensible engineering choices that make the system production-grade rather than a demo.
12 · Honest Engineering Trade-Offs
Every production system has trade-offs. SOP’s are documented in the Master Audit and Review Notes:
Known Risks
- Stealth DOM Brittleness (P1):
xcom-stealth.tsandreddit-stealth.tsrely on DOM selectors that break when X or Reddit update their UI. This is an inherent risk of headless browser automation without official API access. Recommended fix: LLM-powered visual healing or unofficial REST endpoint fallback. - Hippocampus Memory Accumulation (P1): The vector store under
data/brain/appends indefinitely with no TTL or relevancy decay. Over months, this risks lookup latency degradation. Recommended fix: chronological decay or relevancy-weighted pruning. - Dead Code (P2): Several helper scripts in
src/utils/appear superseded by the Intelligence Engine but remain in the build tree. - Test Coverage Gaps (P2): 24 test suites exist, with high coverage on core orchestrator lifecycle. However, Playwright stealth workflows (
test-live-x-threads.ts,test-reddit-interactive.ts) are manual/semi-automated rather than CI/CD deterministic. Recommended fix: mocked HTTP response tests. - Single-Operator Risk: The system is designed for and operated by one person. There is no multi-user auth, no role-based access control, and no ops handoff documentation beyond the setup runbook.
Remediation Roadmap
Implement chronological decay or relevancy-weighted pruning to maintain lookup latency.
Research unofficial REST endpoints or build LLM-powered visual healing for Playwright selector failures.
Audit src/utils/ to safely delete superseded helpers.
Replace live interaction tests with mocked HTTP responses for stable CI passes.
13 · How It Was Built
This was an AI-orchestrated build, not a conventional hand-coded project. The critical skills were not typing speed but systems architecture, constraint design, and quality assurance.
| Dimension | Assessed Level | Evidence |
|---|---|---|
| System Architecture | Architect | War Room, 18-thread model, Ollama/Gemini routing |
| Integrations & APIs | Senior | 9+ APIs with varying auth (OAuth, Bearer, GraphQL, IMAP) |
| AI / ML Ops | Staff | Hippocampus vector memory, Intelligence Engine, multi-agent negotiation |
| Frontend / Full-stack | Mid / Senior | React Vite dashboard with state machine, real-time telemetry |
| Ops & Resilience | Senior | Iron Dome global rate-limit interception |
| Product & Strategy | Architect | VERITAS constraint system, programmatic brand policy enforcement |
Scale: No-code → Hobbyist → Junior → Mid → Senior → Staff/Principal → Architect
14 · Cost & Effort Reality
The BUILD_COST_ANALYSIS estimates rebuilding at 7 skill tiers. All assume 2026 market rates:
The AI-assisted build compressed ~1–1.5 months of active orchestration into what would otherwise require 3.5–16 months. The primary cost shifted from “typing code” to “systems architecture and constraint design.”
15 · Honest Comparison
| Dimension | SOP V6.0 | CrewAI / LangGraph | Agency-Built Custom |
|---|---|---|---|
| Production-ready publishing | 4 platforms, live | Scaffolding only | Depends on scope |
| Content quality gates | 5-layer stack | Not included | If specified |
| Anti-hallucination layer | VERITAS | Not included | Rare |
| Multi-agent consensus | War Room (3 agents) | Task delegation | Usually single-model |
| Hybrid local/cloud LLM | Ollama + Gemini | Single provider | Single provider |
| Rate-limit resilience | Iron Dome | Basic retry | If scoped |
| Vector memory / learning | Hippocampus | RAG patterns | Usually stateless |
| Review → reward loop | End-to-end | Not included | If scoped |
| Operations dashboard | React Nerve Center | CLI/logs only | If budgeted |
| Estimated rebuild cost | ~$100k equiv. | $0 (framework) | $89k–$350k+ |
Key distinction: CrewAI and LangGraph provide agent scaffolding. SOP provides a complete, opinionated, running system. An agency build can match SOP’s capabilities if scoped and budgeted accordingly — at $89k–$350k+.
16 · Who This Is For
Wants to understand the War Room, VERITAS, and hybrid model routing patterns to adapt them to their own multi-agent systems.
Assessing AI-orchestrated development capability. The Skill Scorecard and build cost data provide a concrete evaluation framework.
Exploring whether the solo-builder + AI-agent pattern can deliver production systems at a fraction of agency cost.
Benchmarking: what would this cost my team to build? The 7-tier cost analysis provides direct comparison data.
17 · The Architect
Robert McCullock — Founder & Principal Architect, Design Delight Studio.
Robert built SOP V6.0 as an AI-orchestrated project, acting as the system architect and product owner while directing an AI coding agent. His background spans systems architecture, API integration, AI/ML operations, and product strategy — assessed at Architect level for system design and product strategy by the Skill Scorecard.
- Portfolio: ddsboston.com/pages/robert-mccullock-professional-portfolio-2026
- Email: Robert@ddsboston.com
- LinkedIn: linkedin.com/in/robert-mccullock-8580a8140
18 · Frequently Asked Questions
Free. No paywall. No signup. No certificate. Just the work.
Explore the DDS Vibe Academy
More masterclasses documenting real AI-built systems, step by step, with every prompt and script you need.
Explore the DDS Vibe AcademyThis is an independent technical masterclass. Gemini, Ollama, Shopify, WordPress, Klaviyo, Resend, Playwright, X.com, and Reddit are trademarks of their respective owners and are referenced for identification only; no affiliation or endorsement is implied. All metrics (38,190 LOC, 173 files, 9+ integrations) are verified against the audited codebase as of June 2026. Cost estimates assume standard 2026 market rates and are approximate ranges, not guarantees.
