(1) In numeric/si, the argument "num" should probably be restricted to
finite, real numbers.

(2) The argument "base" of numeric/comma should probably be named "radix".

(3) The reference implementation of numeric/si cannot handle
non-positive numbers, in particular, zero.

(4) In https://srfi-email.schemers.org/srfi-159/msg/8213803, Alex
suggests that the "micro sign" (U+00B5) may be more appropriate than
the "Greek lowercase mu" (U+03BC). However, the Unicode Consortium
says that the Greek letter character is preferred:
https://en.wikipedia.org/wiki/Micro-#Symbol_encoding_in_character_sets.

(5) The reference implementation of numeric/si uses a local variable k
and forces that k <= (vector-length names). However, it has to force k
< (vector-length names), as otherwise formatting 1e30 yields an error.

(6) numeric/fitted of the reference implementation does not test for
precision being zero. In that case, no decimal point should be
printed.

P.S.: I am currently working on an alternative implementation for SRFI
159; that's why I have been finding all the minor issues I have been
reporting recently. (Actually, writing an independent implementation
for a SRFI looks like a great way to verify it.)