[Data-modeling] Historical currency

[Jeff Thompson]

Christopher R. Maden wrote:
> we should be
> able to capture the distinction without overcomplicating the model — or indeed without knowing the
> dates involved. This last point is a problem with time-valued properties; if I know that something
> was true in the past but I don’t know *when* it stopped being true, I can either lie (excuse me,
> “guess” — er, “estimate”) or make an assertion that appears to be currently true; neither is ideal.

The problem assumes a property can only be true over one continuous
time span.  But think of a country occupied by the Romans, then beaten
back, then occupied again, then finally defeated.  What is time-value for
the property that says the country has Roman currency?  Time-valued properties
need to allow for multiple periods.  So in your case above, if you know that something
was true in the past, even for a day, then assert that period.  If
someone else can show that it was true for some period after that, then
that period of validity can be added too.

Only asserting for a day?  Why does this seem counter-intuitive?
If something is true during a time period, does that mean it is not
true after the time period?  Not necessarily.  This is the
dreaded "frame problem":  Just because you know X is true
under condition Y does not tell you whether X is true or false under
conditions outside of Y.  We have naive default assumptions one way
or the other in various circumstances, but logically it's a "problem".
http://en.wikipedia.org/wiki/Frame_problem

- Jeff