For real numbers, precision is checked during limit computation from difference of two adjacent approximations. Unfortunately, this doesn't always produce correct result, since two approximations can be closer than their inf and sup would give as precision. It's possible precision needs to be tracked separately from the approximated value.
Also lim inf and lim sup operations are not implemented.

These are standard functions, should consider proper way to impmement them.

instance Optimal (Complex Double) is only partially implemented in SIMD.lhs.

There are new SI unit prefixes quetta, ronna, ronto, quecto. Should implement support.
https://en.wikipedia.org/wiki/Metric_prefix?wprov=sfla1

Math.Number.Units

Bug in Math.Number.R.RClosure Floating instance.

I think foldMap for the linear Identity map doesn't work correctly, should apply the function.

the fromLinear_impl function doesn't work if the map is the generic identity map special case.

document generation seems to produce links to dependency graphs that are broken.

Check if isospin values are correct.