[rak1507]

There's a difference between one line and short/terse/elegant.

  {m:(x,x)#til x*x; r:til[x]-x div 2; 2(flip r rotate')/m} 

generates magic squares of odd size, and the method is much clearer. This isn't even golfed as the variables have been left.

[gabiteodoru]

Do you know which method is this? Euler's by any chance? And do you have an idea how one would prove that it creates a magic square? It's actually one of my inspirations for writing this, the relationship between the code that does something and the proof that the code actually does what it claims. I'd argue an LLM would find the proof helpful if it were asked to generalize an existing function in some way

[gabiteodoru]

Oh nice solution, thanks for sharing!