Dawid-Skene consensus for multi-agent and multi-LLM disagreement: learn each agent's reliability and the true answer with calibrated confidence. Zero dependencies, node-free, TypeScript-first.

Crowdsourced Triplet Similarity Comparisons 2021

Multi-Agent Reinforcement Learning framework for collaborative label aggregation on noisy classification datasets. Includes DQN agents, Gym-style environment, evaluation tools, and majority-voting baseline.

All-inclusive aggregation rule for multi-label speech emotion recognition: keep every annotator rating instead of collapsing to a consensus label (IEEE SLT 2024)