I don't know if anyone cares about this sort of stuff or not but here are some odd experiment results I've made: I used AutoHotkey to record a successful tree skip into a replayable macro script, but replaying the script didn't produce consistent results. The character only performed a successful skip 6 out of 20 times. Suspecting that AutoHotkey may be a little too slow for the high percision execution, I've written a logger that logged the timestamps of when input is being sent.

By default Limbo outputs 60 fps hard-locked by the vsync. If the rendering and the internal game logic are in sync, you have a window of 16.(6) milliseconds to send your input. The measured deviation of AutoHotkey on my machine was a range of 0.002 to 0.00009 milliseconds so in the worst possible case input was delayed by two microseconds. That figure is way too small to possibly cause such drastic inconsistency so something else must be wrong.

I'm not sure what is up with Limbo but it seems that anything that involves physics adds arbitrary randomness. Very very small bits of randomness that aren't immediately apparent. Like pushing and pulling objects doesn't result in them being in the exact spots every time. Often this isn't a problem for humans because we adjust out actions based on what we see and if the adjustment is as small as a fraction of a second we may not recognize that we even did it but input scripts fail if 100% determinism is not guaranteed.

I'm not very experienced with these sorts of things but I'd like to throw it out there that there may be an undefined behaviour somwhere in the physics code of Limbo that introduces these inconsistencies. In that case there can be no guarantee that the same input results in performing a tree skip from one attempt to another.

Are you still researching the possibility of this? Did you ever manage to confirm it possible, or have you discovered a better option?