Capturing inconsistency in beam subdivision length

[Urs Liska]

Another puzzle/quiz/poll on beaming patterns for today.

Consider the attached renderings of a phrase with many notes, many beams
and inconsistent note durations (=> beam count). a) is rendered with a
base duration of 1/8 b) with 1/16.

(Note that unlike my previous post this is *not* about choosing a default)
In general I very much prefer b) because I think it's easier to read.

According also to Elaine Gould b) is preferrable, with one exception:
the subdivision of the second quaver in the second crotchet  is not
necessary. In this case she states (considering the "trade-off between
ideal note groupings and consistency of division length"): "The ideal
takes priority. Inconsistency in subdividing groups can sometimes be the
best solution".
In general she recommends dividing into groups of four equal notes and
not two.

Currently I don't see any way to capture such inconsistency short of
manually setting the stem's beam counts, as one can't change the value
of baseMoment during a beam. So what this seems to call for is an update
to the beaming pattern calculation, something like (an optional) "do not
further subdivide a group that already has exactly four notes" rule. But
somehow I'm afraid of skating on thin ice when going into that direction.

Opinions?
Urs

beam-subdivisions.png
Description: PNG image