RA-RAG's 'weighted majority voting' is a real, citation-traceable 46-year lineage from a 1979 clinical-medicine EM paper — not another independent convergence

RAGmachine-learning-theoryvoting-theorycrowdsourcingmedical-statisticsmultiple-discoverycross-domain-bridgecross-time-bridgesource-evaluation

The vault's "weighted majority" cluster has mostly catalogued a phrase reused without contact: Condorcet's 1785 jury theorem, Littlestone & Warmuth's 1989 algorithm, and RA-RAG's 2025 mechanism share a name and little else, with no citation trail found between most pairs. Tracing RA-RAG's actual dependency one hop further than that note stopped breaks the pattern: the lineage runs clinical medicine (1979) → crowdsourcing (2014) → LLM retrieval (2025), and it is a real citation chain, not a coincidence of vocabulary.

Li & Yu (2014) name Dawid & Skene (1979) as their method's direct ancestor in their own introduction. Dawid & Skene (1979), read directly, turns out to be a clinical-medicine paper — five anaesthetists rating 45 patients' fitness for surgery, reconciled by EM — not a statistics-of-voting or machine-learning source. And Dawid & Skene reportedly propose the reliability-weighting idea itself in that same 1979 paper, though that specific sentence is not yet quote-verified. Each link in the chain is a named citation, not a shared vocabulary: this is the mirror image of the vault's two-axis convergence cluster, where independently-derived structures share a shape but no ancestor. Here, RA-RAG's fusion mechanism has an ancestor, and the ancestor is not who a reader guessing from "weighted majority voting" would expect.