correlation

What goes wrong if you sort x and y separately before a regression?

Sorting each variable on its own breaks the pairs, so the fit looks excellent whatever the real relationship is. It predicts new data worse than the mean.