Inputs that differ from training data, often causing unreliable model behavior. ← On Chain Proof of Authority →