derivation_id: COUNT(DISTINCT): 7508610 n_distinct: 4336644 (~57% of the true value)
derivation_output_id: COUNT(DISTINCT): 5539406 n_distinct: 473762 (~8% of the true value)
If you expect the ratio of distinct of derivation_output_id values to be roughly linear going forward, you can set a custom value for n_distinct on the column (currently seems like -.0194, aka distinct count of derivation_output_id divided by reltuples of the table). You could also do this analysis every month or six and set the custom value as needed.