I think you may have forgotten to define “regions_input” and “regions_output” layers? They are necessary because in this case the input and output regions of the rule are different.
Since it feels a little silly to effectively duplicate the input and output layers of the rule, I’m considering to derive the input and output regions from the input and output layers in case no explicit “region” layers have been defined.
Yes, the map you sent does work. My tiles are from the same tiles set. My map is 250x150, I don’t know if that matters. I did have a bug the other day where autotiles would only work on part of the map until I restarted. In this situation, restarting doesn’t help. Is it possible I am missing (or have defined) a map property that is preventing it from autotiling. Just to clarify, I am not using autotile while draw, that is OFF and I’m just using the AutoMap (A) command.
So, the problem was a typo in the layer name of the “input” rule.
I think it would be very reasonable if Tiled would report any “input” or “output” rules that are somehow standing alone as errors, since that never makes sense.
Hrm, while looking at implementing this, I came to the conclusion that it would not help in this case. If you define a rule with input_collison and output_collision, the rule makes perfect sense and Tiled has no way to determine there is a typo.
If there is no input layer or output layer at all then Tiled already shows an error message. I’ll also add an error message for when there is no output layer at all.