Ensure you're tracking whether the populace is taking additional measures that are not mandated, and suddenly it starts to explain the error almost completely.
Remember, policies do not execute themselves, and lack of policy does not imply lack of action.
Other problems are also caused by wrong, simplistic models. E.g. SEIR might have estimated the total death count correctly, but not the timing or location at all. (Because the model is isotropic.)
Plugging in annualized data for various airborne viruses into a Bayesian anomalous diffusion model gives much closer timings. Though there are a few models with varying base assumptions and the estimator used a mixture. (One using Spanish flu, next using common cold, another using seasonal flu. Separately the worst case is... common cold variant.)
Remember, policies do not execute themselves, and lack of policy does not imply lack of action.
Other problems are also caused by wrong, simplistic models. E.g. SEIR might have estimated the total death count correctly, but not the timing or location at all. (Because the model is isotropic.) Plugging in annualized data for various airborne viruses into a Bayesian anomalous diffusion model gives much closer timings. Though there are a few models with varying base assumptions and the estimator used a mixture. (One using Spanish flu, next using common cold, another using seasonal flu. Separately the worst case is... common cold variant.)