"The proof of the pudding is in the eating."

Most users want to build a model that will perform proper under any circumstance. Build it right and you don't have to "worry" about it; just go right to the output. This thread is a good illustration of the gnashing of teeth associated with this "right model" approach.

I believe, and promote, a simpler approach but it takes discipline. I think it best to take a good, quick shot at the model and allow the output to confirm or reject the input. Here, if I assume there is an axial restraint through the U-bolt (the moment restraint is a greater reach but follows the same tack), I would add the axial restraint and run the job. If the results (for certain load cases) show large axial loads, I would assume the model inadequate (for those load cases). If the load is low, no big deal.
The opposite is true. If I ignore the axial restraint in my model but I see large deflection (or large rotation for that moment restraint), then I might want to adjust the model.

In most cases, then, the anxiety over the "proper" model is greatly reduced and you can spend your time worrying about the really hard stuff.

The down side of this approach is the requirement that each of these simplifications must be evaluated before other results are examined. Many do not have this discipline.

Good recipies do not necessarily make good puddings. It's the results that count. Focus on the results.