Universal Intelligence Architecture

Thesis and 2nd Order Vision

The core thesis

MasteryMade is a venture studio that partners with proven domain experts through a 50/50 JV model. We extract their intellectual property through a systematic 8-module framework, deploy it as AI-powered systems, and compound revenue across a growing portfolio. Every new expert partner makes the system smarter, faster, and more valuable for every other partner.

This document is not a product roadmap. It is the operating system for a business where agent swarms extract, deploy, market, and monetize expert IP with minimal human involvement. The human role shifts from execution to orchestration — identifying the right experts, validating extraction quality, and making strategic decisions about portfolio composition.

The compounding portfolio effect

Each expert deployed teaches the system. Brad's TIGER QUEST extraction revealed patterns in how sales methodologies get structured. Those patterns made the Bridger extraction faster. Each extraction compounds the institutional intelligence of the pipeline itself. The same applies to content intelligence — cross-portfolio pattern recognition is a moat no single-expert platform can replicate.

The dogfooding loop

MasteryMade is the first expert deployed on its own system. Our skills, content intelligence, extraction pipeline, and daily briefs are all products of the same architecture we deploy for JV partners. When a prospect evaluates partnering with us, they experience our system in action. The Reveal demo isn't a slide deck — it's the system extracting their content, analyzing their competitors, and generating assets in real time. The demo is the product.

The "download Samuel" paradigm

The future of MasteryOS is not a SaaS with a static sidebar. It is an agent that knows the expert's methodology, knows the user's context, and generates a dynamic experience per user per session. When a user "downloads" Align360, they get Samuel's thinking as an agent that operates across their preferred channels and adapts to what the user needs.

Effects cascade

1st: Extract one expert, deploy one clone, generate one revenue stream. 2nd: Pipeline gets smarter with each expert. Content intelligence compounds. Tools we build become the demo that closes the next partner. 3rd: Agent swarms handle extraction, deployment, marketing, and monetization autonomously. 4th: The portfolio of expert agents becomes an ecosystem with cross-expert synthesis.

Org Registry Architecture

The org registry is the skeleton everything attaches to. Read-only shared map. Every service, skill, agent, and workflow references it to understand what else exists, how things connect, and where to find detailed documentation.

Design principles

Read-only reference, not a gatekeeper. If the registry is down, services still work. Living and self-healing. Temporal traces on every change. Lazy-load pattern. Registry stores summaries; detailed docs live in each service's MANUAL.md.

Components

Services manifest: Every service/skill/agent registered with name, capability summary, schemas, status, owner, MANUAL.md path, HC page URL, dependencies, changelog. Entity graph: Every expert, competitor, niche with gate assignment, related entities, ingestion status, extraction progress. Contracts table: API contracts between services — input/output formats locked here before implementation.

HC Protocol as registry interface

The org registry is published as an HC Protocol-compliant page — living web page with machine-readable JSON metadata, human-readable content, and embedded agent that can answer questions about the system architecture. Extends the Neural Registry pattern at jasondmacdonald.com/knowledge-registry.

Supabase schema

Gate Architecture

The gate determines what content enters the system, how it gets tagged, and what retrieval scope applies. Gates prevent personal RSS noise from polluting clone training data and prevent competitor intelligence from being auto-injected into irrelevant contexts.

Universal Content Ingestion Service

One service ingests every source type. Not separate scrapers — one service that eats content from any source and stores it as structured, annotated raw material in Supabase. What changes isn't the ingestion — it's the lens applied after.

Standard output schema

Every ingested piece produces: content_id (UUID), source_type, source_url, gate (1-4), entity_id (FK), raw_text, annotations (JSON — Gemini visual, hooks, format tags), metadata (JSON — engagement, date, author, duration), embedding (pgvector 1536-dim), ingested_at, status (raw | annotated | scored | processed).

Research Service (pre-demo intelligence)

Sub-service that auto-runs when a prospect enters Gate 2. Takes company name/URL, produces: company info, tech stack, recent news, pain points, audience demographics, channel presence, content cadence, competitive positioning. Feeds The Reveal and pre-call briefs.

PRD reconciliation

Three previous documents unified: RSS Intelligence Router PRD (Feb 2026) → absorbed as RSS/YouTube extractors. Content Notebook PRD (Feb 2026) → n8n pipeline becomes the scheduler. Signal Engine concept → Gate 1 configuration with Signal Engine lens applied post-ingestion.

Suggest-Don't-Inject Engine

The nervous system. Content gets stored automatically, but connections between content only become retrievable after explicit approval. This is the difference between an intelligence system and a noise machine.

Three layers

Layer 1 — Passive edge detection (automatic): Runs on every ingest. Compares embeddings within SAME gate. Writes suggested_edge if similarity exceeds threshold. Types: same-topic, extends, updates, contradicts.

Layer 2 — Cross-gate detection (scheduled): Daily scan for connections ACROSS gates. Higher scrutiny flag. Does this competitor's ad reference the same framework as this expert's transcript?

Layer 3 — Architect agent review: Three outcomes: Approve (edge permanent, pattern strengthened), Suppress (marked noise, pattern weakened), Defer (revisit later).

Approval thresholds

Learning loop

Every approved edge strengthens the pattern that generated it. Every suppressed edge weakens it. Temporal traces mean every edge knows when it was suggested, approved/suppressed, confidence level, and who decided — so the system's judgment is auditable.

Expert Extraction Pipeline (8-Module Framework)

The core IP. The systematic process for extracting an expert's intellectual property — not just what they say, but how they think, decide, teach, and what makes their methodology unique.

Pipeline position: expert-research → EXPERT FACTORY (this) → expert-os-deployment

Process

1. Gather source material (dense content first). 2. Run Module 1 — create thinking rubric. Validate before proceeding. 3. Run Modules 2-8 sequentially, validating each against Module 1 rubric. 4. Run Module 9 — retrieval logic. 5. Three-pass validation: forward, backward, ground truth. 6. Store in Supabase as expert_chunks with vector embeddings.

Gate 2 → Gate 3 validation

Prospect (Gate 2): extraction on public content only. Signed (Gate 3): private docs shared. Compare Gate 2 extractions against Gate 3 ground truth. This improves the pipeline's accuracy on public-only content over time — making Gate 2 pre-deal demos more impressive.

Extraction history

Intelligence Lenses

Same ingested data, different analytical output. These don't build clones — they generate intelligence for decisions, content, and strategy.

Competitor Intelligence lens

Gate 4 content → positioning map, gap analysis, ad hook taxonomy, audience overlap, pricing comparison, content cadence. New skill needed: competitor-intel-analyzer (references Composio competitive-ads-extractor and Merci Larry Meta Ad Library workflow).

Marketing Rubric lens

Gates 2+3+4 content with engagement metrics → scoring rubric for hooks/formats/CTAs/body structure, winner/loser pattern matching, format effectiveness per niche, rubric refinement. Key innovation: synthetic audience testing against ICP avatars from expert extraction. Skills: blue-ocean-insight-engine, jason-writer, linkedin-growth-engine.

Signal Engine lens

Gate 1 content → daily intelligence briefs filtered against goals/rocks/sprints, execution directives with falsifiable win/loss conditions, knowledge graph updates. Current: Daily Brief v3.0, Briefs #47-51+, Neural Registry at jasondmacdonald.com/knowledge-registry.

Personal Alignment lens

Gate 1 context → 321 thread management, domino identification, anti-drift checking, morning/evening rituals. Skills: 321-exec, 321-framework-jason, jason-guardian, morning-kickstart, session-close, 2026-execution.

Output Systems and Dynamic Experience Generation

Dynamic Experience Generation

Replaces the static MasteryOS sidebar. Agent reads the user, surfaces what's relevant, generates the interface dynamically. HC Protocol agentic tier — the page itself IS the AI assistant. The agent is the product. The channel (web, Telegram, WhatsApp, voice) is just delivery.

Distribution and Syndication

Performance feedback loop: Engagement metrics re-ingest into universal content service. Rubric scores performance against predictions. Rubric improves. Next batch is better. The flywheel applied across entire portfolio.

MANUAL.md Convention and HC Protocol Publishing

Every service/skill/agent has a MANUAL.md. Universal convention, provider-agnostic. Any agent that enters the ecosystem gets one bootstrap instruction: "Read MANUAL.md first."

Contents

Identity: Name, version, owner, purpose. Capabilities: What it does, input/output schemas, edge cases. Dependencies: What it calls, contracts expected. Failure modes: Known breaks, recovery methods. Changelog: Every correction, timestamped, with "learned from" references. Temporal traces: Version history showing evolution.

Self-healing loop

Services update their own manuals. New failure mode → logged. User correction → logged. Over time, each manual becomes a comprehensive knowledge base written by the service itself.

Runtime-specific thin pointers

CLAUDE.md → "Read MANUAL.md. Apply Claude-specific behaviors." GEMINI.md → "Read MANUAL.md. Apply Gemini behaviors." The manual is permanent institutional knowledge. The runtime config is swappable.

Dual-format publishing

MANUAL.md in repo/file system. Build step publishes as HC Protocol page to web. Both in sync. File-based agents (Forge, CLI) read MANUAL.md. Web agents (NowPage, MasteryOS API) read the HC page. Same content, two interfaces.

Runtime Environments

Forge session logging

Closing the archive gap: every Forge interaction → session ID → full transcript to Supabase → entities/topics extracted → optionally publish to NowPage. n8n workflow handles lifecycle. Once this exists, Forge becomes primary for everything.

The Reveal and JV-in-a-Box Pipeline

Complete prospect-to-partner pipeline. Every section above converges here.

Phase 1 — Identify targets

Proven domain expert, existing audience, content library, monetizable methodology. Sources: warm network (Derek), Athio inbound, Signal Engine discovery.

Phase 2 — Pre-deal intelligence (Gate 2 + 4)

Agent swarms: Research Service runs. Universal ingest scrapes public content. Competitors identified and ingested. Extraction pipeline runs on public content (lower confidence). Marketing rubric scores content vs. competitors.

Phase 3 — Build pre-pitch assets

NowPage go-giver playbooks. MasteryOS v0 (agent swarms stand up basic clone from public extraction). Prospect-specific analysis pages (Zeus Jones / Deloitte pattern). Expert-domain daily briefs as prospecting tools.

Phase 4 — Execute The Reveal

Pre-call: Research Service already ran. On call: voice agent listening, real-time transcription. Live: NowPage publishes page during call. Share link: "[prospect].nowpage.com" — they open it, embedded agent active, they ask a question, agent answers using their methodology. Close: "When can we start?"

Phase 5 — Gate 3 transition

Expert shares private docs. Gate 2 extractions validated against ground truth. Full 8-module extraction runs. Clone improves from v0 to v1.

Phase 6 — Deploy and launch

v1 deployed. Landing/sales pages. Lead magnets from frameworks. Content machine templates (Mastery Labs matrix). Promotion calendar. Social syndication. Three-wave launch: Inner Circle (5) → Early Access (50-100) → Full Launch.

Skills and Services Inventory

PRDs to Reconcile

Parallel Build Lanes

Not sequential sprints. Parallel workstreams referencing the org registry, building simultaneously, registering outputs on completion.

Lane A — Forge / Jason

Universal ingest service. Suggest engine. Session logger. Daily brief agent. Extraction agent. Batch MANUAL.md generation.

Lane B — Sumit / Dev Team

Align360 betaap.io (ship to Samuel). Key Vault. Stripe subscriptions. Voice agent gating. Hidden admin plan. Claude Code on codebase.

Lane C — Lee / Claude Code

Supabase schema (registry, ingest, suggest). API contracts. Embedding pipeline. n8n workflows.

Lane D — Agent Swarms

Expert extraction agents (Align360, The WAY, Brain Muka, Chris Comer). Competitor scraping. Content generation. NowPage publishing. MANUAL.md + HC page batch generation for 24+ skills.

Coordination

Every lane reads registry before work. Writes to registry on completion. Shared Supabase tables need contract agreement FIRST. Registry is the seam.

Mastery Labs Content Matrix

Content creation workbook and promotion calendar at labs.masterymade.com. MasteryMade dogfoods first, then every JV gets the same system loaded with their content.

Integration: Marketing Rubric → content briefs → workbook. Expert Extraction → voice/frameworks → content in their style. Competitor Intel → gaps → content fills them. Promotion calendar auto-generated → distribution routes execute it → performance feedback → rubric improves.

Viral loop rule: Every teaching/email sequence includes a required step where users engage others. Homework equals distribution.

Revenue sprint learning: Cold LinkedIn outreach <1%. Pivoted to warm network + content-led. Content machine generates warm-approach content: NowPage playbooks, expert-domain briefs, personalized analysis pages.

MasteryOS 6-SILO Reconciliation

Lower-Order PRDs — Linked Build Documents

Each section expands into its own detailed PRD at plan.jasondmacdonald.com. These contain implementation specs, Supabase schemas, service architectures, skill definitions, and sprint-level build instructions. Every document links back to this registry as its parent.

Items Unlocked — March 25, 2026 Session