In the display case between l-systems and classic fractals like the Mandelbrot
set, we have IFSs. There is a lot of definitional overlap between these things;
mostly the difference is about *where* they are used and the algorithms used to
draw them.

- Why does no-one (seem to) use linearised Gaussian error propagation to choose
smoothing kernel size when rendering? would reduce the amount of iterations
necessary to overpaint (
*I*think it would; but there must be some reason no-one does it) - Is there a meaningful acoustic equivalent?
- At some point I would like to understand how the Barnsley IFS algorithm inferred IFSs from images.
- Construct IFS as a point process / branching processe.