@BertM
From what I've experienced, the route back to the track is calculated as if you were planning a route—and in fact, some programs do exactly this: the track remains unchanged, and an additional route is generated.
Of course, I don't know the exact program logic, but I imagine it works like this:
First, various entry points are determined (i.e., points where a routable path intersects the track), possibly as an initial approximation based on the straight-line distance.
Then the resulting routes are calculated based on an explicit or implicitly assumed routing profile (fast, short, winding, etc.).
The result that best matches the profile specifications is then used to generate the route back to the track
