<p>Federated learning (FL) and split learning (SL) enable collaborative model training while preserving data privacy. However, these approaches are inherently designed to create a single model and involves sharing of sensitive model information like model updates in FL and activations in SL. Consensus Learning (CL) addresses this limitation by combining ensemble methods with consensus protocols without leaking any information regarding the models. While promising, traditional CL implementations suffer from a lack of transparency due to localized confidence calculations. In this work, a Web3-based implementation of CL is proposed in order to enhance transparency and accountability. The proposed approach leverages smart contracts for reputation scoring and IPFS for persistent storage, ensuring verifiable and auditable confidence calculations. A reputation-based weighting mechanism is incorporated as a fallback strategy to handle potential failures in consensus protocols or any other unexpected errors. The study utilizes the Snow family of consensus protocols, known for their leaderless and Byzantine fault-tolerant nature. Experimental results demonstrate that the proposed reputation-based weighting scheme outperforms traditional accuracy-based approaches, particularly in the presence of malicious actors.</p>