I've already mentioned that if we take the N-et, and set r = N/ln(2),
and then calculate

r' = (r+G+1/8)/ln(r)

we get an adjusted tuning after setting N' = ln(2) r'. Here G
represents the nearest Gram point, which is round(g(r)), where

g(r) = r ln(r) - r - 1/8

and "round" rounds to the nearest integer. This strikes me as almost
black magic, it's so easy. Another piece of the same magic is this:
define a function

tend(N) = 180 (g(r) - round(g(r))),

where again r = N/ln(2), and the "180" makes tend read out in degrees
from -180 to 180. Tend gives the tendency of an et, being positive
for ets with a sharp tendency, and negative for flat ets. We have for
example:

N tend(N)

7 -23
10 8
12 13
15 42
19 -40
22 22
27 75
31 -22
34 40
41 -11
46 15
53 -3
58 67
72 -55
99 54

When using these to create MOS of M steps out of N, it is better that
the tendencies of M and N agree. Thus 19, 31, and 41 are reasonable
fits to the flat 72, while 22, 46 (and 21, where we have tend(21) =
14) are less apt, and 58 is downright awkward. On the other hand,
when adding two ets to get an et, then it is better if the tendencies
are opposite, where they tend to cancel. For instance both 22+31 and
19+34 lead to the neutral 53, whereas adding the slightly sharp 12 to
the distinctly sharp 15 leads to the very sharp 27.

Both the meantone and the 72 systems tend towards flatness, and it
might be interesting to look to the sharp systems (such as the 15 out
of 27 system I mentioned) for something a little different. 22 out of
46, or 27 out of 58, anyone?

I haven't had any zeta feedback--does any of this make sense?