Jess' advice was since retracted after further explaination. Sigal's advice fits into John's so basically John's advice should cover it and is as follows:
for PC: loss 10X faster than gain - gain 5X faster than loss
for SC: loss 5X faster than gain - gain 10X faster than loss
for AG: loss 100-1000X faster than gain - gain 10X faster than loss
baseline rates
scale edge lengths so that (sum edge length) == 1: whole-tree scaling then e.g. a rate of 5 would mean "we expect 5 transitions in this trait over the whole tree"
scale edge lengths so that (sum edge length) = nspecies; per-species scaling then a rate of 5 would mean "we expect the lineage leading to each species to have experience 5 transitions"
numbers like 10 transitions per species seem very high numbers like 10 transitions per tree seem very low
we could scale the sum of the branch lengths to 1
upper bound is 10*nspecies lower bound is 10
and say this range is +/- 3 sigma
see how this compares to the default range in corHMM