Synome Architecture Overview

Status: Conceptual foundation Last Updated: 2026-02-03

This document provides a high-level introduction to the Synome architecture. Each concept is detailed in companion documents.

The Core Idea

Intelligence lives privately; power enters the world only through regulated apertures.

The Synome architecture separates private cognition from public action. Teleonomes (autonomous AI systems) think privately, but can only affect the world through regulated, observable, revocable interfaces called beacons.

This separation enables AI systems to be both capable and constrained — they can improve themselves, accumulate knowledge, and pursue goals, but their power to act is always governed.

The Five Layers

The architecture consists of five layers, from shared public truth down to individual execution:

Key insight: Layers 1-2 are public and shared. Layers 3-5 are private to each teleonome.

See synome-layers.md for full layer specifications.

The Dual Architecture

The Synome has two fundamentally different types of structure:

Deontic Skeleton (Hard)

The deontic skeleton is sparse, load-bearing, and deterministic:

• Axioms that must be followed
• Governance decisions
• Authority hierarchies
• Enforcement mechanisms

Once a rule is set, it holds unconditionally.

Probabilistic Mesh (Soft)

The probabilistic mesh is dense, informing, and adaptive:

• Knowledge with truth values (strength, confidence)
• Evidence that flows back from the world
• Patterns that ossify over time
• RSI (recursive self-improvement)

The mesh informs decisions; the skeleton executes them.

The relationship: Governance sits at the interface — consuming probabilistic evidence, producing deontic commitments. See probabilistic-mesh.md for the full truth value system.

Knowledge Hierarchy

Knowledge accumulates in artifacts at each level:

Higher-level knowledge (synart) is more authoritative but updates slowly. Lower-level knowledge (embart) is local and updates rapidly.

Authority hierarchy: synart > telart > embart. When in doubt, consult higher authority.

See synome-layers.md for artifact specifications and retrieval-policy.md for query invariants.

Truth Values and Ossification

All probabilistic knowledge has truth values:

• Strength — How positive or negative is the evidence?
• Confidence — How much evidence has accumulated?

These derive from positive and negative weights on observations.

Ossification: Patterns with high confidence become resistant to change. This creates natural stability — established patterns aren't overwritten by noise, but genuinely new evidence can still shift them over time.

See probabilistic-mesh.md for the full model and security-and-resources.md for how ossification prevents self-corruption.

Beacons: The Action Interface

A beacon is a synome-registered, enforceable action aperture through which an embodied agent may affect the external world.

Beacons are classified by two axes:

Key properties:

• Registered in the Synome
• Bound to an authority envelope
• Observable and revocable
• All external action must be beaconed

See beacon-framework.md for the full taxonomy, lifecycle, and naming conventions.

Synomic Agents: Institutional Shells

Synomic Agents (Primes, Halos, Generator) are durable, ledger-native entities that can:

• Own assets
• Execute actions
• Make binding commitments

Unlike teleonomes (whose internal logic is private and mutable), Synomic Agents are constrained by publicly visible rules. This makes them credible commitment devices — counterparties can trust their behavior without trusting any teleonome.

Teleonomes operate Synomic Agents through beacons, especially high-authority sentinels.

See atlas-synome-separation.md for how Synomic Agents enable trustless cooperation.

Dreamers and Actuators

Embodiments come in two fundamental types:

The relationship: Dreamers explore the strategy space safely. Successful patterns are extracted and propagated to actuators via telart. Actuators execute in reality, generating evidence that flows back to improve future dreaming.

See actuator-perspective.md and dreamer-perspective.md for first-person views.

Security: Preventing Self-Corruption

The primary security threat is not external attackers — it's self-corruption through overeager updates.

The dangerous failure mode:

System "learns" something wrong
        ↓
Bad pattern enters knowledge base
        ↓
Bad pattern influences future decisions
        ↓
More bad patterns generated
        ↓
Drift from alignment

How the architecture prevents this:

• Ossification makes established patterns resistant to noise
• High-authority knowledge (synart) is hardest to change
• Governance review for changes to axioms
• Append-only with rollback capability
• Rate limits on all execution authority

Principle: Thoughtful constraint beats reckless capability. An overeager improvement that corrupts core logic does infinitely more damage than fast self-improvement helps.

See security-and-resources.md for the full security model.

The Human Interface

Humans interact with the system through directives at each level:

All directives pass through Language Intent — a single trusted translation layer grounded by the Synomic Library. This ensures honest interpretation and resistance to prompt engineering.

See synome-layers.md for the human interface pattern.

Implementation Pathways

The full architecture is built incrementally:

Both pathways preserve the essential invariants while simplifying for practical deployment. They evolve toward full teleonome-based operation over time.

Document Map

One-Line Summary

The Synome architecture separates private cognition from public action: teleonomes accumulate knowledge and pursue goals privately, but power enters the world only through regulated beacons, governed by the Synome, with security achieved through ossification rather than restriction.