Aegis Insight: Knowledge Graph Infrastructure for Detecting Suppression and Coordination Patterns in Document Corpora

Current retrieval-augmented generation (RAG) systems optimize for information retrieval accuracy while remaining blind to the epistemological structure of information landscapes—specifically, whether high-quality research is systematically marginalized or whether consensus is manufactured through coordinated messaging. Unlike misinformation detection systems that identify false claims, we address fundamentally different questions: whether true claims are systematically suppressed and whether consensus around accurate information is artificially manufactured. We present Aegis Insight, an open-source knowledge graph system that detects three categories of information manipulation patterns: suppression (quality-visibility gaps, network isolation, institutional dismissal without engagement), coordination (temporal clustering, language similarity, synchronized emotional triggers), and cross-cultural anomalies (isolated cultures exhibiting identical complex patterns). The system employs a seven-dimensional extraction pipeline processing documents into typed claims with entities, temporal markers, geographic references, citations, emotional content, and authority-domain relationships. Detection algorithms implement threshold-based "Goldfinger" scoring where isolated indicators score minimally but accumulated indicators trigger exponential escalation. We validate against historical ground truth: Thomas Paine (documented state suppression, 1790s), Smedley Butler (Business Plot suppression, 1930s), and Yellow Journalism coverage of the USS Maine explosion (documented coordination, 1898). Results demonstrate effective discrimination: suppressed figures score 0.78–0.83 (CRITICAL) while non-suppressed controls (Benjamin Franklin) score 0.39 (MODERATE) with zero suppression indicators. Coordination detection correctly identifies the Yellow Journalism campaign with 66 near-identical phrases across sources, 24 claims within a 14-day window, and synchronized emotional triggers (41.7% fear, 30.1% urgency). The system runs entirely locally on consumer hardware via Docker using local LLM inference through Ollama, with checkpointing for multi-day extraction of large corpora. We release Aegis Insight as open-source infrastructure for epistemological analysis, suitable for integration with existing RAG systems via Model Context Protocol (MCP) endpoints.