In my understanding, Randomized Benchmarking (RB) generates a sequence of Clifford gates with different lengths and then characterizes the average error. Since RB is not sensitive to SPAM error, it doesn't characterize it.
My question is, does the error rate obtained by RB include decoherence error (I know it includes gate error)? When the length of the gate sequence increase, I understand that the gate error will accumulate. Does the decoherence error influence the final result due to the T1/T2 of each qubit? Also, in RB result, there is a constant when fitting the curve. Is the decoherence effect too small to be accounted for in the computed error per Clifford?