Broken imputation of censored incidence in the ML method ?

To investigate after code cleaning/refactorign has been conducted.

For now, this method seems to accumulate estimated cases (if they are right?) just before the first observed data. This is not right.