HP calcs are really not that accurate..

[AlexFekken]

Inspired by some other thread (where I looked at the pros and cons [but mainly the cons, the pros getting enough support already] of QPI and PSLQ), I just realised that we can "deal with" some of the issues raised in this thread in a more sophisticated and more defensible way if instead of sneaky extra digits we use (auxiliary, stateless) CAS and functions like QPI and PSLQ behind the scenes.

Would we still be talking about "lying" then, i.e. in particular when the auxiliary CAS does not know the provenance of the floats? Presumably you would need a mechanism to indicate how accurate your floats are...

More generally, how useful, magical, dangerous, perfect, .... would such a CAS-backed numerical calculator be?