This chapter argues that esteem for police experience has not yielded the cumulative changes in policing expertise and sensitivity to changing context that allows complacency. It suggests that the move towards cumulatively evidenced policing should start, and it's most feasible route lies through the respect for, but refinement of, lessons learned via policing experience itself. Research evidence is neither personal nor vicarious experience. It is metaphorical cub killing and political infrastructure of policing which makes the accumulation of knowledge about what works best so imperfectly realised. In proposed policing application, Bayesian statistics simply formalises and makes public what the more skilled officer does already. A refined version of this approach satisfies the criteria of prediction based, iterative, scalable and accessible. The acquisition of skill consists of a process of predicting likely outcomes, observing the actual outcomes and revising conditional probabilities.