AU-Vic_SW.2HS-1E as an example.
What appears to be happening is that the final position of each x/y/z coordinate in the child model is correctly translated by -250/0/600 (as specified by the Include statement in the parent model) but then also an additional incorrect translation of -61/0/3 as specified by the ARM statement in the child model. This second translation by the ARM statement should not occur.
Then when the component is rotated 45 degrees, it uses the point -250/0/600 + -61/0/3 (= -311/0/603) as the point of rotation, which is correct, but since the component is in the wrong place to begin with, the rotated component is still in the wrong place.
So the rotation bit seems to be working ok, just the initial include is erroneously using the point of rotation as an additional factor in the translation.
The correct operation is seen in the editor and v103, but not with 105 for all my semaphore models - see screenshots.