and probably wouldn't be picked up via replay checks.
Replay checks would absolutely be unable to detect it. It would need to be done via manual examination, at which point the need for an experimental to test it is very little - sure, the exp could confirm 100% for sure whether there's an issue or not once a suspect location is identified, but it wouldn't help in spotting places where there's
likely to be an issue.
If we can get a rough idea of how many
rough-terrain or
containing-steep-slopes levels use Jumpers, that might be a useful starting point. If this is very few, it may be the case that even if
all of them were to be affected - keeping in mind "affected" would be "new backroute is introduced" rather than "intended solution breaks" - it wouldn't be an issue. On the flipside, if there's a lot, it might be safest to avoid making the change just in case.
I'm also somewhat inclined, seeing as this only really came up as an issue in theoretical discussions prior to the Jumper's release, to say that nothing needs to change because it's proving to be fine as-is.
There are basically three possibilities here regarding the overall outcome:
Option A - a special case for a pixel 2 above and 1 to the right (or left, if facing left) of the lemming's starting position, similar to the special cases for builder->jumper at 0,-1 or for dehoister at 0,0. Should have very little breakage, but introduces yet another special case.
Option B - explaining this requires understanding how the Jumper moves - each frame is divided into multiple 1px horizontal or vertical (not diagonal) movements. So for example, on the first frame, the Jumper moves (1px each step) up, up, forward, up, up, forward. The change would be - if a lemming's foot check detects a pixel,
and the next movement would be an "up", the lemming only transitions to another state if the pixel 1px above his foot position is
also solid. Current behavior remains if the next movement is forward or downwards. This could have wider breakage, but an advantage is that it removes the builder special case (as this new general rule would cover it).
Option C - leave it as is. My preference at this stage, simply because it hasn't proven to be a major issue.