When doing some aggregations to compute the amount of slots per unit of time (like in the queries: query_slots_per_second.sql & slots_by_second.sql), the slots per unit of time is assessed by doing a sort of average of the slot usage ("SAFE_DIVIDE(total_slot_ms, )").
This makes the (false) assumption that the slot usage is always the same during the whole duration of the query.
Instead of using the JOBS_BY_PROJECT tables, it is possible to use the JOBS_TIMELINE tables, which give a better insight of what happens each second