I with I could have stated everything in your post as well as you did. The only part of it that I take small exception to is the above quote. It is true technically but in practice, failures caused by model misspecification typically represent the limits of our knowledge and force us to average over things we don’t know such as risk factors that were omitted from the model. Other approaches just cover up the problem. See Incorrect Covariate Adjustment May Be More Correct than Adjusted Marginal Estimates – Statistical Thinking