neatokeen!

It's such a fantastic feeling when

1) Your simulation doesn't work;
2) you do an integral by hand to surmise why (biased MCMC samples due to a naive lognormal jump kernel);
3) you do 5 more pages of not-completely obvious integrals and approximation
4) (having a happily large number of terms cancel out!);
5) to solve for a reparametrized correction;
6) and change your code to use this method;
(here's where it gets good:)
7) and finally, it actually solves (or at least greatly mitigates - it's not clear whether a proper Solution even exists) the problem!

You know, I don't get nearly enough of step 7) in my life. Sadly step eight remains elusive:

8) Someone authoritative gives a damn about it.