Expectations are model consistent, on average, in a RE framework, so errors are made, but on average they are zero.
In a rational expectation framework, do all agents know the true law of motion of the economy? But they make errors, only on average, they are correct, right? So, the average agent does know the law of motion then? But this agent does not exist... . Well actually the average agents equals the individual agent, because all agents are the same, representative agents? But how then do you explain the random errors with zero mean?