The inbred diallel is one of the oldest designs in model organism genetics. Describing the complete (or incomplete) set of F1s produced from a given panel of parental lines, it can be used to cleanly partition different types of heritable effects, from additive, epistatic, parent-of-origin, and sex-specific versions thereof. Since pilot data often corresponds to an incomplete diallel, diallel analysis of such data can powerfully guide selection of follow-up studies. Yet, thanks to a turbulant history marked by racorous disagreements, awkward formulations, and outdated statistical ideas, diallel data is often not analyzed as such -- indeed, diallel analysis is too often treated as an arcane puzzle to be avoided. In our view, this wastes a tremendous opportunity. Motivated by a series of extensive pilot studies on the founders of the Collaborative Cross, we recently revisited diallel analysis from a modern statistical perspective, reformulating the traditional model and approach to leverage what we see as some of the diallel's greatest strengths -- the ability to project into a new design space in order to prioritize follow-up experiments that use genetic combinations of the same parental lines. We have since extended our approach to define and estimate heritable effects on response to an applied treatment, leading to a definition and modeling of GxE that would be almost impossible in human studies. I will first introduce basic concepts of the diallel cross and its decomposition via Bayesian hierarchical modeling; then I will describe how these are extended to model genetic effects on treatment response, illustrating this extension with recent work on understanding the pharmacogenetics of haloperidol treatment, and the immunogenetics of response to flu infection.