News and archival information in games

I enrich the typology of players in the standard model of games with incomplete information, by allowing them to have incomplete "archival information" - namely, piecemeal knowledge of correlations among relevant variables. A player is characterized by the conventional Harsanyi type (a.k.a "news-information") as well as the novel "archive-information", formalized as a collection of subsets of variables. The player can only learn the marginal distributions over these subsets of variables. The player extrapolates a well-specified probabilistic belief according to the maximum-entropy criterion. This formalism expands our ability to capture strategic situations with "boundedly rational expectations." I demonstrate the expressive power and use of this formalism with some examples.