In QEC, to evaluate logical error rate in numerical simulations, we usually do noiseless syndrome extraction at the end of the protocol and decode and apply correction based on the noiseless syndrome. This leads to logical error rate $P_L=O\left(p^{(d+1)/2}\right)$ . However, in magic state cultivation , when they simulate up to cultivation stage (without escape stage), they seem to do noiseless syndrome extraction and post-select using the noiseless outcomes at the end of the protocol, which le