[STKFLT]

I don't have any additional insight to the format, but I think the idea is that there is an implied ->infinity on every date range. Every bank can only have one bank_name so multiple bank_names for the same bank entity can be sorted on the 'valid' and 'recorded' axes to find the upper bounds of each.

[dragonwriter]

In the bitemporal model, both system and valid times are half-open intervals, and both the preceding and following interval can either have a different value or no value. Using only start times means that while records can be updated in either time stream, they cannot be logically deleted (in transaction time) or invalidated (in valid time) once they exist. There are databases where this assumption is valid, but in general it is problematic.