It all started with "there is no good 3D analogue of eth Mandelbrot". At some level, this is understandable, because the easiest way of seeing the 2D Mandelbrot iteration is as "squaring and adding" and a 3D vector multiplied by itself is, well, nothing.

But, not really happy with this, I started thinking about what "squaring" and "adding" actually meant, geometrically, in the complex space. Squaring (well, all multiplications) are simply "rotate and scale" and adding is simply translation.

With this in mind, I spent some more time thinking what would be needed. Specifically, we'd need a 3D transformation matrix, based on the "last step", followed by a translation (this being the point in 3D space that we're interested to see if it belongs to the set or not).

A translation is simple, just add the point/vector that is the result of the last transformation to the point we are checking for membership in the test.

This is a bit more complicated. First, we define taht we're using a right-hand coordinate system. Then, we generate a transform matrix based on the previous poiint, the new point, and teh last transform.

1. Align a transform matrix so its first ("X" axis) is the latest point.

2. With this done, we can then define that we want the transform to be "in the XY plane" of the last translation. This, basically, means we take the cross-product of the previous point computed and the latest point computed. This gies uz the "Z" axis of out transform.

   However, we also want the next transform to be as close as possibel to the existing transform, so we also compute the negative of said Z axis, then pick the closes of Z and -Z as our new Z axis.

3. We can then generate the remaining axis of our transform matrix as the cross-product of our X and Z axes.

4. To get the scaling right, we then forcibly rescale the Y and Z axes to be the same length as the X axis.