Done, I think. I also tried to make it look a bit better.

I'm pretty sure I've swapped them around now.

I've changed it so that for the random cases, it randomly decides to either do no errors (20%), a message error (40%), or a parity error (40%), and have combined the larger and smaller tests into one batch.

I mostly am doing many tests because an answer which is almost correct might only fail in a small number of cases, and I wouldn't want slightly wrong answers to pass just because the case where they fail didn't come up in the testing.

I recently had to change how I did it to resolve one of the issues raised, so the examples now look like:
Test.assert_equals(correct(2,3,'11111010001'), '11111010001')
Test.assert_equals(correct(2,3,'11011010001'), '11111010001')
Test.assert_equals(correct(2,3,'11111011001'), '11111010001')
Test.assert_equals(correct(2,3,'11111010011'), '11111010001')
So it is '111110' + '10'(rows) + '001'(cols). laid out it looks like:
111 1
110 0
00 1
I think your looking at the previous versions of the examples, which made sense then because I was doing columns and rows the wrong way around.

You're right. I believe I've now swapped the M and N to be correct, but it required many changes so I may have done it wrong :/

I've added another batch of random tests that only modify the parity bits. Does that work?

That's some outstandingly terrible code there! Well done.