We can add ‘wobble’ to the list of causes.
With enough wobble, a cog collides with the rail either on the outer edge or within that square cutout. It can also get hung up on ice chunks, even if it doesn’t reach the rail.
The drive wheel also has some significant (full quarter turn) gear lash before the drive train engages, which means any time the tracks engage, one track will drive before the other and the core will yaw off-course by a significant amount. But for now I’ll count that as a separate issue.
I’m hoping that the design engineers did really, really well, and it’ll just be a bearing or something simple to maintain. We’ll find out when the ticket is replied to!