In 2008, I wrote a semi-technical paper entitled "What makes a good role model" on the occasion of Jim's 75th birthday. The paper had to be dropped for reasons that will become apparent, but it succeeded
in pleasing Jim who liked conceptual papers with simple technical content.

I will introduce the role model strategy and show, to my immeasurable distress, how its solution reduces to elementary Monte Carlo integration in the non-parametric case. Its relevance to the parametric case was not immediately obvious because parametric estimation is not normally considered in discrete probability. I will show an example involving the
constraint node operation in a belief propagation SUDOKU solver where the role model strategy is used successfully.