@cstross typically fatuous assumptions at work here. why should one assume that it takes much more information to describe a million sins than to describe only one? sins generalize with ease. that's why the concept of cardinal sins exists, because sins fall readily into a compact set of abstractions, and all individual sins boil down to "being A (the sinner) willfully injured being B (the victim)" with the rest being incidental detail in the eyes of a Creator who already knows all details of Creation anyway.

@mxchara But sins are committed by sinners! Which implies more than one bit of information per sin. Indeed, there may be mitigating or modifying parameters. Every instance of a sin is unique and must be accounted for separately!

@cstross I suggest that there's great compressibility of information in this case. how much distinct information is really needed? aren't all murders basically the same crime (cf. Cain murdering Abel)

@mxchara @cstross Dante already described arranging different types of sinner in different parts of hell. So run length encoding should be quite efficient

@bassthang @mxchara But nobody has just ONE sin! Everybody has one OR MORE sins. So you actually need a vector to encode their sin, one that can accommodate both the nature of the sins and their severity separately. To compare sinfulness of the undeserving one then needs to consider homogeneous matrix transformations …

@cstross @mxchara Apart from a few exceptions, most people only have a few vices. So the matrices will be sparse. We have good algorithms for that. I think it's computationally possible. We need to do some prototyping ... (... reaches for some cocaine, hookers, lube, and the NumPy documentation...)