I used R graphics to visualize what my own solution was doing, and to make the images that appear in the kata description.

As a hint at the time I solved this, the console output is not sanitized, so you can output SVG graphics to both see the input and verify your output.

Fixed it :D
Now you have to return None in such cases

Alr raised as issue

Ahaha great :) No problem, and thank you for enjoying it :)

(Accidently posted twice)

Sorry for the late reply (I haven't been active on Codewards), but no, unfortunately this kind of approach will not work.

ah, I never saw this message before because you put a spoiler flag on it x) (no notifications in that case).

about the number of valid/invalid, it's not a problem if it's not balanced. Moreover considering the number of base nets isn't that big.

and thx :)

"Comparing with a reference solution would mean I have to reduce the amount of test cases or find/use a better solution than mine as a reference."

Sorry, am I missing something here? You shouldn't have to work backwards. Start with an n by m random grid, and play the game forwards. Then give the result of the grid to the predecessor function, possibly cropping on any side where it's possible. (The dimensions of the problem could be anywhere from 0 to n+2, and likewise for m, depending on whether it blows up or shrinks in any direction.) The downside is that you might have to be more selective, computing a few extra random scenarios and throwing out or fixing the stuff that withers away too quickly. But because you derived each problem from a legitimate source, you know that there's at least one solution.

👉😺👉 Righteous! 🙌