Thank you @schuemie sounds like we have a known problem when there are more then one record per subject_id. The default behavior of FF is to use subject_id as row_id. So, given a situation like this
We have a problem using FF default behavior of rowId = 'subject_id', because we are more likely to have difficulty differentiating between features generated for the same subject_id with different cohort_start_date 5/1/2016 and 2/15/2017.
In this case, we need to use a structure something like this where we need to create a new column that uniquely identifies every row record within the same cohort_definition_id
where rowId = 'cohort_row_id' . Current standard tools don't do this by default, and cohort table does not have cohort_row_id. So we have to do it outside - by creating a new rowId field by using row_number() (partition by cohort_definition_id order by subject_id, cohort_start_date)