ARIC-FL: An Atomic Reputation-Aware Incentive Contract Framework for Blockchain-Enabled Federated Learning

Abstract

Federated learning (FL) enables distributed data owners to train a shared model without moving raw data to a central server.However, practical FL still faces an incentive and governance problem: clients spend computation, communication, energy, and privacy budget, while the task publisher cannot directly observe the true quality, reliability, or cost of each participant.Existing mechanisms based on Shapley value, contract theory, auctions, reinforcement learning, reputation, and blockchain smart contracts usually optimize separate parts of the incentive pipeline.This paper reviews representative incentive-driven and blockchain-enabled FL studies using six criteria: contribution evaluation, participant selection, reward allocation, privacy protection, attack resistance, and deployment cost.The comparative findings show that fair valuation remains computationally demanding, selection rules require independent contribution validation, reputation introduces cold-start bias, and blockchain improves auditability while adding transaction overhead.They also show that privacy-preserving valuation and atomic reward settlement are rarely combined with budget-aware participant selection in one lifecycle.Based on these findings, the paper proposes ARIC-FL, an Atomic Reputation-Aware Incentive Contract framework that combines reputation-aware reverse auction selection, adaptive contract menus, privacypreserving contribution scoring, validator commit-reveal auditing, encrypted off-chain model exchange, and smartcontract escrow settlement.A review-based evidence matrix and gap-to-design traceability assessment evaluate the framework at the design level, while a reproducible protocol is retained for future empirical validation under IID, non-IID, and adversarial settings.