Lineage-Aware Retrieval-Augmented Generation (v5): Prisca-GraphRAG with Sparse-Autoencoder Feature Interpretability
Stephen P. Lutar · Zenodo (CERN European Organization for Nuclear Research) · 2026
Paper v5 of The Ouroboros Thesis. Presents Prisca-GraphRAG (lineage-aware knowledge retrieval), VOTE-RAG (Omega-weighted Borda-count ensemble), Federated Prisca Privacy (Bekenstein-calibrated differential privacy), and Tawa SAE (ceque-radial indexed sparse autoencoder interpretability). Reference implementation in TypeScript. Part of the SZL Holdings governed AI platform.Part of the Ouroboros Thesis chain by Stephen P. Lutar Jr. (ORCID 0009-0001-0110-4173, SZL Holdings). Reference implementation: github.com/szl-holdings/ouroboros. Source repository: github.com/szl-holdings/ouroboros-thesis.