I opened up your file and ran an optimization. I can confirm that your RMS spot radius gets significantly worse after optimizing. If you look at your merit function editor in the column “% Contribution” what you’ll see is that 99.9% of your merit function value is coming from one operand:
 

MNCA is the minimum center thickness of air between two surfaces. You’ve set a minimum value of 0.1mm, and the current value is -0.094, meaning that there is a surface with a negative thickness. Let’s find which one…

This thickness isn’t even a variable, it’s one that you’ve set. So let’s first undo the optimization to get back to your starting point, update the surface range for MNCA from 1-11 to 2-11, and re-optimize. I optimized for only 45 seconds and then stopped it, to see if we’re heading in the right direction. I can now see that the spot size has in fact reduced as expected, and that the boundary operands are not contribution to the MF value at all.