Mark, ya, I agree it's most prudent to let sleeping dogs lie.

In the one "real" complaint we got (issue 24546) the cause was never determined - but double rounding was ruled out in that specific case, and no _plausible_ cause was identified (short of transient HW error).

As the years go by, my interest in "doing something" about a bug nobody has apparently encountered in the wild, and remains theoretically possible on an ever-diminishing number of platforms ("32-bit Linux not using faster SSE arithmetic" seems to be the universe), has faded somewhat  ;-)

So I'd close as "won't fix".  If I had to do something, I'd add different implementations under an "if this box does double-rounding" import-time test so nobody else has to pay for it.

But I'm cool with whatever Raymond settles for.