I wonder if this topic will be subject to a review of sorts (wrt the oscillator being open for business) as you can see here, it can momentarily catch you out when switching machines
Further to that it’s quite nuanced in that even when you kill the oscillator’s dedicated decays, the sound does not drop as the machine may not have been triggered yet
The suggestion of linking envs will improve some of those issues where you may see/expect the voice as being silent already and unwittingly create those instantaneous transients
For overlapping notes there may be issues with mitigating this tendency
However it does sound like an idea they could explore to cover your scenario and it may be that the solution for the former and this one can be arrived at together so that the voice slips into a managed ‘silent’ mode