We analyze the evolution of the COVID19 pandemics and show that the basic compartmental SIR model cannot explain the data, some characteristic time series being by more than an order of magnitude different from the fit function over significant parts of the documented time interval. To correct this large discrepancy, we amend the SIR model by assuming that there is a relatively large population that was infected but was not tested and confirmed. This assumption qualitatively changes the fitting possibilities of the model and, despite its simplicity, in most cases, all the time series can be quite well reproduced. Nevertheless, in some cases (i.e., countries or regions) the estimated susceptible population decreases too fast. In such a case, the observed dynamic is only due to the transitions between the two infected compartments--the confirmed infected and the unconfirmed infected--and the rate of closing the cases (by recovery or death) in the confirmed infected compartment. Our analysis proves that the number of infected people is significantly larger than the one recorded and we provide a method to estimate it. We also discuss some relevant extensions of this model, to improve the interpretation and the fitting of the data.