First thing, those links to SRFI 142 and SRFI 141 still need to be swapped.

Second thing:

It's my understanding that one of the goals of this SRFI is to permit very minimal and highly efficient compiler output for calls to these API.

What happens when bit-set? is called with a bit index that is a fixnum but is greater than fx-width?

There is a mathematically correct result, and returning it does not violate the fixnum rule. But in order to properly generate it the code must check for this condition.

The question is also raised by every other function where a bit position, bit-range or shift amount are given. The notion of a "bit-field" referring to the sign-extension part of a fixnum doesn't even make useful sense.

It would be helpful for generating fast code if all such functions made it an error to refer to a bit position >= fx-width (or > fx-width for an "end" position). Similarly shift amounts should be restricted to the non-inclusive range (-fx-width,+fx-width).

Then a compiler can blindly generate shift instructions and trivial fixnum patch-ups without having to check that the shift amounts are >= the register width -- for which the fast code would emit a mathematically incorrect answer.