Determining the correct data model underlying TSQL will probably be
the most difficult of all the tasks. Unfortunately, and not
coincidentally, this task is a central one, on which most of the other
tasks are predicated. To focus the design, I advocate that time be
added to the data model in two separate steps, with the first to
add valid time and the second to later add transaction time.
A proposal is needed that confronts the
controversies currently raging in the research community, including
1NF versus ¬1NF, temporally grouped versus temporally ungrouped, tuple
time-stamped versus attribute value time-stamped, homogeneous versus
non-homogeneous, events versus intervals, interpolated versus stepwise
constant data, recurrent events, and whether keys should be required.