Topological Truth: Conflict Resolution Through Knowledge Graph Structure

When conclusions are recorded as nodes in a knowledge graph with typed edges between them, the graph develops structural properties that can resolve contradictions — when the structural evidence is sufficient — without external judgment. This paper presents a conflict resolution system that classifies and resolves contradictions using only the topology of supporting edges — no large language model, no human judge, no voting mechanism. LLMs construct the graph; the resolution phase is pure topology. Applied across three experiments at increasing scale (236, 894, and 1,336 nodes), the system auto-resolved 16–67% of conflicts. Manual review of 20 post-resolution conflicts revealed that 55% were construction artifacts, while the resolution mechanism correctly classified every one as requiring human review. Cross-source analysis revealed empirically distinguishable structural signatures between within-source and cross-source edges.