Dawn Models: Post-Symbolic AI Architectures

---

Overview

Dawn Models implements post-symbolic AI architectures based on Dawn Field Theory principles. The primary model is GAIA v2 — a modular intelligence architecture where specialized modules compose via a PAC conservation bus. Also includes TinyCIMM domain-specialized variants, SCBF interpretability framework, and CIMM Legacy.

This is part of the Dawn Field Theory ecosystem.

GAIA v2 — Modular Intelligence Architecture

GAIA v2 treats intelligence as a composition of specialized modules — language, reasoning, safety, memory — connected by a conservation bus that enforces PAC conservation at every boundary. The bus uses SEC entropy phases for zero-parameter routing and RBF balance for self-regulation. Built on Fracton SDK 2.1.

Modules:     Language | Reasoning | Safety | Memory | Observability
                 |          |          |        |           |
Bus:        [  PAC conservation  |  SEC routing  |  RBF regulation  ]
                 |          |          |        |           |
Substrate:  [              Fracton SDK 2.1                          ]
                 |          |          |        |           |
Interfaces: [ Spinal Column | MCP/Agents | GRIM | Kronos Vault     ]

Key properties:

• Conservation as contract — PAC violation at any boundary = detectable hallucination
• Zero-parameter routing — SEC phase of input determines which module handles it
• Continuous learning — O(1) per token, no retraining, no catastrophic forgetting
• Modular composition — swap, add, remove modules without retraining

Validated Foundations

Finding	Confidence
PAC conservation holds (residual = 0) across all experiments	High
O(1) learning per token, zero gradient descent	High
12.5x memory savings via delta-only PACTree	High
100% cross-model embedding graft (GPT-2 to Pythia)	High
Hallucination = +9.6% PAC violation in GPT-2	Medium-High
12,000x MLP advantage with Mobius neurons	Medium-High

TinyCIMM — Domain-Specialized Variants

Five lightweight models implementing PAC/SEC/MED principles for different domains. Self-contained, no cross-dependencies. Boltzmann and Mobius are ancestors of GAIA v2's safety and reasoning modules.

• TinyCIMM-Euler: Mathematical pattern recognition with 6-metric SCBF instrumentation
• TinyCIMM-Navier: Fluid dynamics with turbulent breakthrough detection (4/4)
• TinyCIMM-Planck: Quantum-inspired adaptive architecture with grow/prune
• TinyCIMM-Mobius: Continuous learning via Mobius transformations (12,000x MLP advantage)
• TinyCIMM-Boltzmann: Hallucination detection as PAC violation (+9.6%)

SCBF — Interpretability Framework

Symbolic Collapse Bifractal Framework — measures symbolic collapse and bifractal patterns in neural network weight evolution. Standalone analysis tool.

CIMM Legacy — Production Engine

Mature entropy-based intelligence engine with Bayesian optimization, multi-agent consensus, and superfluid dynamics. Apache-2.0 licensed for production use.

Repository Structure

dawn-models/
├── research/                 # AGPL-3.0 — Experimental
│   ├── GAIA/                 # Modular intelligence architecture
│   │   ├── src/gaia/         # v2 source (core, modules, interfaces)
│   │   ├── tests/            # v2 tests
│   │   ├── spikes/v1/        # Archived v1 (25 POCs, 8.4K lines production code)
│   │   └── .spec/            # v2 spec + v1 spec
│   ├── scbf/                 # Interpretability framework
│   └── tinycimm/             # 5 domain-specialized models
├── stable/                   # Apache-2.0 — Production
│   └── cimm-legacy/          # Production CIMM engine
├── roadmaps/                 # Development plans
│   └── gaia-v2-roadmap.md    # GAIA v2 milestone plan (M0-M8)
└── docs/                     # CONTRIBUTING.md, LICENSING.md

Getting Started

# Install Fracton (required for GAIA v2)
cd ../fracton && pip install -e .

# GAIA v2 (in development)
cd research/GAIA
pip install -e .

# TinyCIMM variants (self-contained)
cd research/tinycimm/TinyCIMM-Mobius
pip install -r requirements.txt

# CIMM Legacy (Apache-2.0, production)
cd stable/cimm-legacy
pip install -r requirements.txt

Licensing

Use Case	Location	License
Academic Research	/research	AGPL-3.0
Open Source Project	Either	Respective
Commercial Product	/stable	Apache-2.0
Specialized Commercial	/research	Contact us

See LICENSING.md for complete licensing strategy.

Dawn Field Theory Ecosystem

• dawn-field-theory — Core theoretical foundation
• fracton — Computing SDK (GAIA's physics substrate)
• reality-engine — Physics simulation and validation
• dawn-models — AI architectures and implementations
• dawn-infrastructure — Deployment and ops

Contributing

See CONTRIBUTING.md for guidelines.

Contact

• General Inquiries: info@dawnfield.ca
• Research Collaboration: info@dawnfield.ca
• Commercial Licensing: info@dawnfield.ca

See LICENSING.md for complete licensing strategy.