> 'fxxx' = confidence level xxx where
> xxx is a three digit integer 0-100
> representing  percentage reliability.
A 100% reliability is probably impossible. We could limit that to two 
digits (01 to 99 percent).

> I have been using "questionable" while
> others are using "uncertain", and we
> should settle on one or the other.
The word "questionable" may be interpreted as if different people do not 
agree, whereas "uncertain" is more like "nobody knows". Maybe this 
distinction may be useful.

> the letters a, b, c, d, e,
I would prefer letters that cannot be misinterpreted (by humans or 
filters) as hexadecimals. It is more unambiguous (and not more difficult) 
to write g, h, i, j, k in the spec and implement it.

> 1) I think the default is
> "unstated reliability".
It seems good.

I also wonder about the asymetry of the reliability. What about an event 
that happened "probably 1910, maybe 1909, and maybe as late as 1978"? 
Then, the center of probability (1910) is not at all the arithmetic mean 
of the span. Maybe a simplified form of parametric probability density 
could help.

> conversational predicates..
What do you mean?