Stockfish doesn't backtrack as far as I know. It's more likely that Stockfish is recalculating the position and is unable to find the same fastest mate as before, even when the new calculation is done at a higher depth (due to resources etc.).
It's highly unlikely that Stockfish will find a forced mate at a certain depth and later adjusts it because it found a refutation at a higher depth. You can safely assume that once a mate in X is found, it will not be replaced by a worse mating sequence.
Feel free to post an example.
In rare occasions, the distance to mate can increase as Stockfish evaluates further. This happens without resetting the evaluation or clearing the cache. As I understand, this was supposed to be impossible. When Stockfish announces "mate in N", isn't this a forced mate? If later it finds a longer forced mate line, it should simply ignore it. Or just backtrack as soon as it reaches a higher number of moves. What am I missing?