Cognitive Version Control: Fact, Concept, and Assumption Ledgers for Long LLM Conversations
A lightweight specification and prompt toolkit for reducing premature convergence and belief-revision failure in long LLM conversations. It combines three ledgers (facts, concepts, assumptions), conversation forking, and three measurable metrics: concept reuse rate, frame entropy, and belief revision latency.
Authors
- Dongze Li
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-18
- DOI
- https://doi.org/10.5281/zenodo.24338017
- Primary Topic
- Speech and dialogue systems
- Type
- article
- Field-Weighted Citation Impact
- 0.00