Dynamical modeling for the spread of an infectious disease has 100 years of history. Traditional compartmental models such as SIR and SEIR, empirical models and data-driven models were all used during the COVID-19 epidemic. They played an important role in advising policy makers on whether social contacts should be suppressed or whether it is better off waiting for herd immunity to run its course. The predictions by these models however yield dubious results. Here we revisited the model formulation from first principles, and show why they failed.