Question about PRD within LAMMPS

Dear all,
This is a question on how dephasing step is handled within the PRD routines of LAMMPS.

Ideally, if an event is detected during dephasing step, the corresponding replica should reset to original configuration and repeat the dephasing step using a set of newly generated velocities. This will ensure that the forward dynamics of the system is accounted during other stages where the clock is advanced (MD, or correlation run).

Steve/Ray/Axel et al. please comment on the above.

Thanks in advance,