FractalVideoGuard v0.5.2: Production-Ready Deepfake Video Detection via Fractal-Informational Ontology (FIO) and Long-Range Dependence (LRD)

FractalVideoGuard is a production-ready deepfake video detection system based on Fractal-Informational Ontology (FIO) / QO3 Universal Attractor Theory and long-range dependence (LRD) analysis. The pipeline extracts interpretable features from video streams and files, including DFA-based Hurst exponent estimates, fractal dimension (box-counting) on edge fields, and frequency-domain artifact proxies (DCT/FFT blockiness/ringing). Release v0.5.2 focuses on production safety and reproducibility: (1) hang-proof rotation metadata detection with timeout safeguards, (2) memory-stable ROI standardization via buffer reuse for large batch processing, and (3) numerically robust ringing proxy with MAD flooring and overflow/NaN guards. The configuration system provides validated, type-safe parameters (presets for high_quality/fast/debug/mobile), JSON/ENV serialization, and deterministic runs via explicit seeds and full config capture. Recommended use cases: forensic analysis, large-scale batch screening, and near real-time moderation (config dependent). Inputs supported: local video files, webcams, and network streams (RTSP/HTTP) when supported by OpenCV backends.