Don:
this is indeed an intriguing question, because there are
certainly no rules how the slope of beams should be chosen.
In the case you described that led to the ambiguity, I would predict that
90% would chose a small negative slope to emphasize the
descending line, despite the fact that the height of the
last note is substantially higher than the first one.
I am not sure whether this small differences can be formulated
by an algorithm. It is obvious that small corrections must
be always made, because the computer can not be taught to understand
music. Nevertheless, I have the impression that the 1-note differences
should have a larger weight: since three of them are negative and only
one of them positive (regardless of its large value 5),
in ambiguous cases the trend of the majority should decide to tip
the balance, in this case towards the negative value.
Christof
