Entity Integrity is a practice definition by Lee Sharks concerning identity preservation in AI knowledge graphs and retrieval systems. It addresses cases in which generated summaries confuse similarly named entities, split one entity into unrelated fragments, misattribute work, erase differentiation through generic categories, or preserve an obsolete identity snapshot.
The document proposes a diagnostic procedure built around the Encyclotron. The procedure maps collision entities, evaluates whether an entity remains coherent across query types, traces attribution, and checks whether the returned description is current. It then specifies a repair artifact containing structured entity data, explicit comparisons with likely collisions, negative tags, Semantic Integrity Markers, and consistent descriptions across multiple public surfaces.
The Lee Sharks entity map is used as a worked example, with Lee Sharkey and Lei Yang as collision risks. The broader significance of the practice is that it treats disambiguation not as a one-time database correction but as a maintained retrieval architecture. Entity Integrity is positioned alongside Retrieval Forensics, Compression Diagnostics, Retrieval Architecture, and the Metadata Packet for AI Indexing.