Pressing skip causes the XT to recalculate all sections of the route, rather than just the sections affected by removing (skipping) one of the route points. Selecting Closest Entry Point also recalculates.
A quick aside. Have you ever loaded a track and selected 'Go!'. Its worth doing to see its behaviour. The track stays in place always, but shows as a megenta line. if you deviate from it, you get a dotted magenta straight line between you and the closest point of the track. It constantly changes as you ride. The bottom right data display shows how far away the closest point is. No navigation just information - "your original track is 3 miles in this direction". Try it. It is surprisingly relaxing.
Another quick aside. If you turn a track into a trip on the XT screen, you get spoken, turn by turn directions. There are no route points, so if you deviate, the XT does the same trick of finding the closest point - but his time it navigates you to it. Part of your route has now changed - there is a new section between you and that closest point. And this becomes the route.
Back to the plot. It seems that when a normal route is recalculated after a skip, It rebuilds the route in a similar way to the last example. If you then deviate significantly - eg turn off at 90 degrees - it will try to turn you round to go back. Nothing unusual there. That is to be expected if it is the faster way to get to the next route point. But it seems that it is not trying to get you to the next route point anymore, it is trying to get you back to the closest point, and it has just laid a new bit of route behind you to get to that first closest point. The new closest point is now behind you. And each time it lays a new bit of route, always will be. Until you get close to the original section of the route.
I don't know this for certain, but I have evidence that strongly suggests that this is the case.**
But all of the above does not happen if the route has been built on the XT screen. It only happens if it has come from Basecamp, MRA, or some other external source.
The XT distinguishes between different types of route - 'Saved' and 'Imported'. They are listed separately when you open up the Trip Planner App. Only the routes created on the XT show as 'saved'. All of the others are imported.
It turns out that in the .System folder is a folder for 'Trips'. each trip file contains some information relating to each trip or route. One of these bits of info is a 'flag'. A boolean true/false value called mImport. If you change this from a 1 (=true) to a zero (=false) , the XT sees this as a saved route rather than an imported route. And the route then behaves perfectly - ie if you deviate, it calculates another way to get you to the next route point. Initially this may want to take you back, but unlike before, it does not persist.
I have been in conversation with tech support about this for about 2 years, providing evidence, and in the last couple of weeks this solution which came about from a discussion as my testing started to home in on the precise circumstances when it happened.
FrankB found the relevant flags. Someone else spotted that the XT separates imported from Saved. I had never seen that, as I always cleared out my XT when starting a new test, and I always use Basecamp or MRA.
------------
*** That evidence - if you turn off U turns, the same behaviour still exists, but instead of u turns you get instruction to turn round using housing estates, side roads, complex road junctions. The important thing is that these can be seen on the map, and each new turn round command results in another loop on the XT's screen. If you turn round and follow the route that it is proposing, it navigates you through all of the housing estate, side roads and complex road junctions where it attempted to turn you round before. Which I think proves that it is laying new sections of route behind you rather than trying to get you to the next route point.
Whatever - they have everything they need. The technician at last recognises the problem, which they cannot simulate, and the solution is actually quite convincing in that it points the finger to the fact that the problem exists.
It would help if Garmin got everyone to use the same terminology consistently instead of introducing a Trip as a new thing for a term that they have already used for something else !