Currently when using the cumulative function, the data is evaluated and returned based on the 'as ordered by the dimension, using hash values for string based fields'.
This returns intuitive results for continuous variables, like dates, and for many ordinal variables, like Time Spent per Visit (Bucketed).
But it also creates scenarios in which unexpected results might occur or the function is less usable.
As an example if a client uses a week range of Monday-Sunday, the data is presented as expected when ordering by dimension, but when using the cumulative function the values are aggregated in a different order as Sunday has a lower hashed-id than Monday.
Additionally for non-ordinal values the function doesn't really have utility - but could! This request would be to instead accumulate the data based on the order of the rows within the table. This would especially be valuable now that table builder provides the ability to display non-ordinal or continuous values based on user selection, rather than metric ordering.
As an example, a classified eVar, in which it would be most beneficial to display data related to categories that have order, but cannot be defined as such within Adobe's back-end.