So, there’s no jitter if you do no board the train?
There may be something there, but that’d be an issue with the CMC perhaps? I wonder if the observed result is due to the inconsistency of spline point distribution. You could try rewriting it using this instead:
Here’s the gist: