I know I have seen the answer to this on this forum, but can't find it.

We always try to model anchor to anchor. But, what if you are only looking at a section: anchored at one point tied into a larger header that goes on forever with some intermediate restraints, which doesn't need to be completely modeled.

I recall a rule of thumb was to include 20 pipe diameters past the last restraint then put lateral restraints at the "free end". Is that reasonable?

The rule of thumb is: "in each direction on the header, go far enough to cross three restraints". Typically this is sufficient, but you do have to evaluate the results, every model is different .

I don't believe there's any real unified method.

Ideally you want to emulate that connection point's displacement in real life. Just ask yourself if whichever rule of thumb you're adopting is likely to achieve that effect were you to model the whole thing in CAESAR.

My preferred method is to attempt to estimate centroids of thermal expansion - i.e. locations where displacement in piping could be reasonably estimated to be 0 - and put the appropriate restraints there.