Performative Personalization Incentivizes Truthfulness in Federated Learning

We study collaborative learning with strategic clients who may misreport oracle outputs to steer the learned model. Under simultaneous realizability and leave-one-out identifiability, we show that a leave-one-out consensus mechanism prevents harmful unilateral misreports and identifies the deviator. We also propose a one-shot alternative that uses a minimal cluster-recovering personalization oracle and preserves incentive compatibility without an $M$-fold increase in computation.