I gave a talk on Bayesian workflow in Stan. It includes recommendations on defining “what” problems to solve and “how” to solve with one’s and community’s resources. This would help you to classify the concepts on model misspecification, coding error, inaccurate computation. I am particularly interested in simulation-based calibration where the model+algorithm as a whole is tested and updated based on the simulated parameters’ posterior coverage. Lecture note:
Comment is the energy for a writer, thanks!