Sovereign Causal Graph: A Neuro-Symbolic Architecture for Air-Gapped Causal Knowledge Discovery with Application to Bitcoin Security Analysis

I present the Sovereign Causal Graph (SCG), a neuro-symbolic system for extracting validated causal knowledge from unstructured technical documents in air-gapped environments. The architecture addresses two critical challenges in enterprise AI deployment: the hallucination problem inherent to Large Language Models (LLMs), and the data sovereignty requirements of regulated industries.

The core contributions include:
(1) The Foss Hallucination Gate - a 14-step deterministic validation pipeline achieving 99.9% rejection of malformed extractions
(2) The Foss-UQA Protocol - systematic extraction of quantified uncertainty bounds from natural language
(3) The Foss Generator - an LLM-free system producing 100,000 synthetic training samples in ~10 seconds
(4) A neuro-symbolic exploit discovery architecture validated through differential testing against Bitcoin Core v0.16 vs v28.0

Evaluated on 525 Bitcoin Security documents, the system extracts 4,309 validated causal triplets achieving 82.6% quantification coverage. The system's focus on P2P protocol divergence was calibrated through technical feedback from Bitcoin Core maintainer Pieter Wuille (SIPA).

This dataset includes the complete pipeline database (4,309 triplets) and the technical whitepaper describing the methodology.