> I think that in 278.sld the definition you give of fl-least-normal is
the value of fl-epsilon.

Yes it is. That's a dumb mistake. I've replaced it with

     (define fl-least-normal
       (do ((candidate fl-least (fl* 2.0 candidate)))
           ((flnormalized? candidate) candidate)))

which seems to do the trick (on binary systems, I don't know of anyone
with a decimal float Scheme).

 >  The Gambit test suite has minimal unit tests for some routines in
this SRFI in this directory:

Thanks, I will take a look.

-- Peter McGoron