Sorry. I mixed several threads into one. It wasn't immediately clear that the second bug isn't related to handling negative values in scale_delta. It's related to gradual time desynchronization between cores, so they should probably be addressed separately.
Thanks.
