The Mirror Fractal Method
A unified mathematical framework for hierarchical binary decomposition. Protected by US Patent Pending #64/034,974 covering eight inventions across compression, protocols, and AI memory.
Core Principles
Recursive decomposition at each level, enabling adaptive encoding and streaming-compatible decompression across domains.
Bit-exact round-trip guarantee. Every dataset encodes and decodes to identical bits, suitable for archival and forensic applications.
No randomness, no training, no model checkpoints. Identical input produces identical output every time, every device.
Exploits sparse structure inherent in events, weights, and discrete geometric objects. Compression improves with sparsity.
Runs on CPU, GPU, FPGA, and neuromorphic chips. No exotic math, no specialized accelerators required.
Single unified framework across all products. Same core primitive powers compression, quantization, adaptive streaming, and cryptographic protocols.
What the Method Enables
Lossless Compression
Achieve 10–50× compression on sparse binary datasets without neural networks. First shipping product: MFC for event-camera data.
Efficient Quantization
Sub-2-bit LLM weight representation with bit-exact accuracy recovery. Planned product: MFQ.
Progressive Containers
Universal file format supporting hierarchical extraction and incremental decoding. Planned: FRAC.
Adaptive Protocols
Streaming protocol that adjusts bitrate and quality in real time based on network and decoder capability. Planned: FQL.
Encrypted Memory
Hierarchical encrypted storage for AI context and episodic memory. Planned: MirrorMind.
Cross-Chain Consensus
Fractal decomposition applied to blockchain consensus mechanisms. Planned: Fractal Bridge.
Patent Status: US Provisional Patent #64/034,974 filed April 10, 2026 (non-provisional filing planned April 2027). Eight inventions are protected: MFC, MFQ, FRAC, FQL, MirrorMind, Fractal Bridge, Mirror Trie Binary Search, Mirror Trie DNA, plus foundational Adaptive-Radix Framework. All patents are jointly assigned to Mirror Fractal Lab and available for licensing.