Why this is the best thing the brain does in public
Most of what a brain does is invisible from outside it. This is not. Visual space is spread across the cortical surface in an orderly sheet, so a patch of tissue you can point at answers to a patch of the world you can point at. Two things next to each other in your view are next to each other in your head.
Turn on how far out and you can see the other half of the arrangement: the centre of gaze takes an enormous share of the cortex, and the far periphery is squeezed into a thin rim. That is why you can read this sentence only where you are looking, and why everything else is a vague impression of shapes.
How the map was measured, and how I checked it
The subject lay in a scanner and watched a bar sweep across a blank field, and a wedge rotate around it. Every point on the cortex has a time course, and for each one a model asked which small patch of visual space best explains when that point got busy. The answer is two numbers per vertex: which direction from the centre of gaze, and how far out.
I did not take that on trust. A real retinotopic map has a property noise does not: it is smooth across the cortical sheet, because neighbouring vertices see neighbouring parts of the world. Measured on this surface, neighbouring vertices differ about thirty times less than randomly chosen pairs, and shuffling the values destroys that completely, scoring 1.00 as it must. The build script runs that test on both hemispheres, before and after the mesh was simplified for the web, and refuses to write anything if it fails.
Each half sees the opposite world
Look at the two hemispheres separately. The left one is almost entirely coloured for the right half of the visual field, and the right one for the left. Measured on this subject: at a 25 per cent fit, 95.8 per cent of the left hemisphere's well-fitted cortex answers to the right of where they were looking, and 97.7 per cent of the right hemisphere answers to the left. Raise the fit quality and it goes to 98 and 99 per cent.
That the numbers get cleaner as the fits get better is itself the argument that these are real measurements rather than noise. Shuffling which hemisphere each point belongs to gives 48.7 and 49.7 per cent, which is chance, as it has to be. The build refuses to write the file if any of that fails.
This is why a stroke on one side of the back of the head takes away one side of the world rather than blurring all of it.
TODO: First1 Last1. There is no paper to cite and nobody
named to credit, which is not how it should be. It came from the NYU Center
for Brain Imaging, OpenNeuro ds007283, and the credit is
theirs.