Mathematically, the term "small" is ambiguous. Perhaps that's why Common Lisp names its corresponding value
MOST-NEGATIVE-FIXNUM.
That said, in GNU APL, these numbers are somewhat bogus anyway. In particular, the actual maximum number that can be stored without loss of precision depends on the underlying data type of the value.
For real numbers, this data type can be either APL_Integer in which the largest number is 9223372036854775808 (2^63), but if you try to create this number in GNU APL using the _expression_ 2⋆63, you will get an APL_Float back, which has a smaller maximum precise value of 9007199254740992 (2^53).
So, in summary. You can never rely on integral numbers being precise to more than 53 bits of precision unless there is a way to force the use of APL_Integer which I don't believe there is.
It would be nice to have support for bignums in GNU APL. It wouldn't be overly difficult to implement I think. Perhaps I'll try that one day unless Jürgen is completely against the idea.
Regards,
Elias