Yep, thank you for testing it.
It’s good to know that a possible workaround is working in a bigger scale. (I know Dave method but in this case I didn’t thought it has to do with the small size of the model), however in this case I think the fastest way is to redo with guidelines as you said.
I didn’t try to figure out why it happens, but doing some tests as you did I got this statements:
- For this value it affects to both sides (as I expected)

- All arcs that didn’t got segmented are composed by 12 segments (I’m working with circles of 48s)

- If I explode one arc and then I compare the segments with the segmented arc this is the difference

- I counted the number of segments, for the bottom one there are 10 segments instead of 12.
Taking in account what you said that it doesn’t happen with another dimensions for the arc It could be what you suggested and it has to do with the start/end segments and the way they merge with the beighbors, but no idea.
Anyway, it’s not a big deal, just curious.