Dear @cryo-lg,
Thanks for confirming the box sizes. It looks like something more subtle may be the cause of this. Could you first confirm that the mask is valued in [0,1] and not some other range? (Just in case – sometimes we’ve seen issues where imported masks have the wrong range).
If the mask is fine, my other guess would be that the particle scales are somehow mis-estimated and that is causing the poor scaling in subtraction. When you ran the previous refinement, did you turn on “Minimize over per-particle scale”? If you’re not sure, you can check the scale distribution in the input particle dataset by running a “Homogeneous Reconstruction Only” job – at the end, it will display a histogram of the particle scales. If scales aren’t all 1, it may be worth it to try re-refining the structure with the “Reset input per-particle scale” parameter on, and then trying subtraction again with the output particle stack.
The other potential cause I can think of is if the window inner/outer radii were changed during any of the processing. This can cause discrepancies with the multiplicative scale, so the subtraction job has to know what window radii were used in the previous refinement to correct the scale (some more info on the subtraction job is here).
Best,
Michael