Note: This program requires a 286+ and an SVGA monitor than can
      handle mode 0x5F (640 x 480, 256 colors).

I've found some interesting effects while using this program to test
the output of raw samples of "randomness" from system sources:

 - raw samples based on system clock drift, when plotted, were limited
   to a small streak on the surface of the sphere with a few isolated
   spots elsewhere.

 - fast timings between Windows message broadcasts passed a variety of
   other tests (Maurer estimation, Chi-squared, uncompressability) yet
   clearly produced a spiral that wound out from the center.

 - timings between keystrokes did not produce a file with as much
   estimaed randomness as above methods yet was as chaotic-looking as
   input from /dev/urandom.

This may be due to the method of sampling those sources rather than the
sources themselves (although detecting flaws in the sampling method is
important).

