The produced output 1.8011300000000001 is at least not what I would expect from ToShortest.
This code path doesn't generate a fallback to bignum calculations. I can't fully deduce from the comments whether this is intentionally.
But the net outcome is that this makes ToShortest a no go for our usage.
We have implemented an overlay of ToShortest that takes an additional max_digits input that we can use to solve the problem, but I'm wondering whether there is a better / more general solution.