Currently we join underestimates and offset using an INNER JOIN, meaning we lose data in case there is no underestimates or no offsets.
These is of little effect on data as it only affects projects with a very small number of uniques (and we know the data quality is lower for uniques <1000 so we recommend it is not used for those). See: https://wikitech.wikimedia.org/wiki/Analytics/Data_Lake/Traffic/Unique_Devices/Last_access_solution#Data_Quality_Analysis