Conflict-Aware Enterprise Retrieval: Preserving Organizational Disagreement Across Heterogeneous Knowledge Sources

This preprint presents Conflict-Aware Enterprise Retrieval, a system-design framework for preserving organizational disagreement across heterogeneous enterprise knowledge sources. The approach extracts structured claims, models relationships such as SUPPORTS, CONTRADICTS, SUPERSEDES, and CHALLENGES, incorporates provenance and temporal state, and uses deterministic answerability policies to decide when an AI system should answer, warn, abstain, or block automation. The work focuses on enterprise engineering scenarios in which approved decisions, implementation state, operational evidence, documentation, and discussion may disagree. The paper proposes an evaluation methodology and benchmark design; empirical implementation results are left for future work.