Constrain one variable to be greater than another

[drbenvincent]

@sammosummo

It’s clear to me why solution #1 produces OK predictions but poor parameter estimates. Essentially there is too much prior density on low values of b (perhaps the prior was a half-normal or something). When b is close to 0, the value of k1 needs to be close to what k2 should be.

Yes, that makes sense in fact. Yep, it was a HalfNormal on b, and looking at the traces, if I had the sigma higher then it could have worked and perhaps masked the more fundamental problem?