I went looking for vals which could be used for approximating the
7-limit miracle canonical map, stepping through divisions of 3 rather
than of 2, and seeing how well the canonical map was approximated. It
turns out that not only does [72, 114, 167, 202] work well, it works
extremely well--vastly better than anything else.
If we do a least squares for (114*x-cents(3))^2+(167*x-cents(5))^2, we
get x=16.68429155 as the size of a step; [7*x, 72*x] gives us
[116.7900408, 1201.268992]
and the difference in cents from the canonical map is inaudible:
[-.6300e-2, -.54236e-1, .37024e-1, -.2922e-2]
We can do the same calculation and get a least squares optimum for
all four images of the primes under the canonical map, with
essentially the same result: x=16.68427953 cents, with generators
[116.7899567, 1201.268126]
and difference from the canonical map of
[-.5434e-2, -.52865e-1, .39031e-1, -.494e-3]