Traditional Byzantine Fault Tolerance (BFT) consensus algorithms effectively tolerate node behavioral faults but lack the ability to verify the quality of input data. This makes them vulnerable to security risks from low-quality or “compliant yet malicious” data in data-driven applications. To address this gap, we propose a Data-Quality-Driven Byzantine Fault Tolerance algorithm based on Zero-Knowledge Proofs, called Q-BFT. The algorithm introduces a “quality gate” prior to classic BFT consensus—an on-chain verification phase that uses zk-SNARKs and is automated by smart contracts. This allows nodes to prove in zero-knowledge that their data meets predefined thresholds for accuracy, completeness, and consistency without exposing raw data. Passing the verification becomes a prerequisite for joining consensus voting. We design a two-layer smart contract architecture that efficiently orchestrates off-chain proof generation and on-chain automated verification. Experiments show that in a 100-node network with 30% malicious nodes, Q-BFT improves the consensus success rate from 41.5% (with PBFT) to 96.4%, while maintaining federated learning global model accuracy above 88%, in contrast to the model collapse (< 20% accuracy) observed under a traditional BFT protocol. The system achieves an average verification latency below 0.65 s and a throughput of 735 TPS(Transactions Per Second), striking an effective balance among security, privacy preservation, and operational efficiency. By enforcing privacy-preserving data quality verification as a mandatory gate before consensus, Q-BFT thus provides a high-assurance foundation for data-sensitive and privacy-critical applications. It addresses the core vulnerability of traditional consensus in scenarios like federated learning, where model integrity depends on participant data quality, and trustworthy data markets, where transaction validity requires assured data authenticity without exposing the data itself.
Li et al. (Fri,) studied this question.