I think CSP’s anti-aliasing is not literally anti-aliasing, but a feature that exists for better results with fill colors on line drafts.
That’s why it seems that transitions are enabled based on the depth of the line edges.
The whole antialiasing you assume in blue above, I can’t think of any reason to need it.
Especially the lineart fills drawn on a single layer with white opacity.
In the parts that need a good transition, it builds the transition, and it doesn’t build the transition where it looks good, I think it looks sharp with 1px at 45% so it’s a logic not to do antialiasing.
(In some paintings, different parts of the line art have different degrees of edge transition, some will look harder, some will look softer, which is used to express texture, etc.)
In other words I think antialiasing should not be really just literal antialiasing, but for good fill.
I also realized that I was being stupid. 
When krita enables anti-aliasing for ordinary line art (the kind with a deep middle of the line), after filling it, it seems that anti-aliasing is not used in some places.
I know it’s not obvious in the whole painting, but it’s more obvious when you look at it alone. (Compared to CSP.) Which led me to post this thread…
I also now find that the spread value adjustment can also improve the fill.
(Actually I hadn’t thought of using it before…)
If the line art is filled after the setting is lower than 100%, the part of the growth selection will automatically generate a transition…
After the MR you submitted is merged, users can also achieve better filling effects.
Finally, thank you for your work and patient explanation.