The most important information that an actor needs is the auragon counts of the other actors; this guides an actor’s decisions in dream combat. But there are two sources of this information, and reconciling them is problematic.
The first is the direct information that actors get during dream combat. This is the actual source of information for all actors. During dream combat, they get an approximate estimate of the auragon counts of each of the other actors. These estimates, however, have uncertainty values, so they might be far from the mark.
The second source of information is the indirect information they get from deals. This information suffers from several possible sources of error: the internal errors of the speaker; the deliberate intent to mislead on the part of the speaker; and the obsolescence of each speaker’s knowledge.
This leads me to suspect that I should reduce the uncertainty of the information gained during dream combat. There are two ways to handle this. With the original Siboot, each actor learned the correct value of ONE auragon count for each of the actors. In my current scheme, however, each actor learns an approximate value of all the auragon counts. Is the earlier system better?
Not really. The older system obviated lies: any statement about an auracount had to be true. Thus, the actor accumulated a set of supposedly absolutely true auracounts that were made uncertain only by the passage of time. In this version, the aura Katsin, which denotes honesty, is meaningful only if actors can have different degrees of Katsin — in other words, can behave with more or less honesty. Therefore, actors must be able to lie, which means that uncertainty is intrinsic to all auracounts.
But this raises all sorts of tricky problems. For example, auracounts are reported in integer format but are actually known in floating point format. If Zubi thinks that Skordokott’s Tanagagon count is 1.87, she has to report it as 2, with some degree of uncertainty. Should her reported uncertainty include the error introduced by rounding off to an integer?
Having established that auragon counts must be floating point with uncertainties, the next question is, by what algorithm do we establish new values? My current algorithm simply adds a new measure in the form x ± u. This is combined with the actor’s current value using the standard mean combination algorithm. But given the obsolescence of information, should it not replace the old value outright?
And should all three auragon counts be presented with equal uncertainties? Would it make more sense to give mixed uncertainties to the different auragon counts?
Here’s another consideration: I think that more Katsin should make lying more successful. But that involves a variety of additional factors.