I would love to see Medium adjust their formatting to make it possible for something like Tables of Contents, or faithful rendering of notebooks that don't just rely on Github gists, but until then, I think it does a huge disservice to the actual information that technical/symbolic-math-including authors are trying to convey.
The information in the Jupiter notebook is different than the one in the article; actually I find math in Jupiter Notebooks also not nice and readable, apart from the occasional equations here and there... for a full calculation it is not the best, even though some extensions help.
I'm generally allergic to tweaking things & general command-line bullshit but the whole process was surprisingly smooth. Readers don't have to deal with popup & tracker bullshit (except for youtube, who have endeavored to make it impossible to embed videos with issuing tracking cookies), and it feels nice to have much more control over what makes it into the final product.
For any quantitatively-oriented person, the quickest solution might be to write the article in LaTeX and host it on github.io, but other solutions are of course possible (e.g. MathJax and some other free web hosting).
Btw, I thought the article was not behind a paywall ... I chose not to profit from it by not activating the option. I need to check again how it works.
For random matrice the author introduces what it means to be rotationally (O) invariant, but is what follow restricted to that? Also it is not too clear why the code is actually doing (creating the same ensemble) as what was talked about before.
This is why is called the Gaussian Orthogonal Ensemble. The Unitary ensemble deals with unitary matrices for example. All of these ensembles deal with diagonalisable matrices.
The code is providing a numerical verification of the analytical formula. It is a simple code that samples symmetric matrices of different NxN size from a Gaussian distribution.
For N fixed, you draw a certain number of samples and you plot the eigenvalue distribution and compare it with the analytics. Then you can see that for small N there is a deviation from the analytical expression, as this was formally obtained in the large N limit.
Hope this clarifies.