In a message dated 96-04-08 04:01:28 EDT, stewarts@ix.netcom.com writes:

Data contains varying quantities of predictablity and unpredictability. Some of the predictability has simple enough structure that a basic compression function can find and exploit it to squash the data. Some of the predictability doesn't. For what it's worth, compressing the data before using it for other things does leave you with somewhat more consistent entropy per byte for "typical" random input, because it eliminates the easy stuff.

That was the entire point of my original posting on this subject. I was proposing using a compression function on spinner data, which contains very little entropy and compresses well. (50 - 80% on idle loop timing data, depending on processor load) I don't believe I said anything about compressing image data of any kind, or audio recordings of humpback whales doing the wild thing, etc. Noise sphere plots of ZIP files look pretty good, regardless of how good or bad the plot of the unZIPed file looks. (Raw idle loop timing plots are terrible.) I have posted a longer reply to Perry via E-mail... Jonathan Wienke