Hi @rj.edwards
Thanks for reporting this - it may be a bug. We are trying to reproduce locally. Is there anything else peculiar about this data that you can think of that might indicate the case where this happens?
Static masking should work as expected with the input mask used at all iterations. For now as you saw, the dynamic masking with similar params to what you used to generate the static mask can be used to maintain a soft mask at every iteration. In any case the default dynamic masking parameters may be too sharp, and we’ll investigate this also.
There is actually a fair bit for us to investigate in local refinement, and your other post Advice for subtraction masks? may be related to this.