I contemplate a refinement in the gossip model. The current model assumes that each character will believe everything. Should there not be some skepticism, especially when implausible things are said?
Should plausibility be a continuous variable or a boolean variable? Suppose, for example, that a character receives a questionable report. Should that character either accept or reject the report completely, or should the character accept it partially? Should emotional reactions be tempered by the trust that the character feels for the informer? Should the character's trust in the informer be diminished in proportion to the implausibility of the report?
Behavioral responses are the province of the storybuilder, so we will need to provide her with a function called Plausibility that assesses the credibility of an event as reported by an informer. But this raises another issue: if something is reported by an unreliable informer, should it not be subsequently reported by a more reliable informer? Gad, that would be a drag on the action. Then there's the whole issue of the TellTree, which assumes a boolean transfer of information.
I think that the way to handle this is to assume full veracity except in those cases in which the plausibility of the report falls below a threshold, in which case the recipient can express skepticism. All of this can be under the control of the storybuilder; the only support I need offer is a new function called Plausibility(Event, Recipient, Reporter), which would take into account the Recipient's trust in the Reporter, as well as the natural plausibility of the event. The problem lies in defining plausibility. It really should be handled through a virtual reaction, but the ideal measure is dependent upon both the change values and the original relationship values. In other words, if Joe does not trust Fred, and Joe receives a report that Fred is untrustworthy, this should be highly plausible. The best measure, then, is the change in Joe's trust in Fred. But I can't actually change the relationship values in this calculation, because those changes will take place further down in the calculation. I suppose that I could simply repeat the internal calculation used in AdjustXXX; that might be the only solution.