This bug report is a joint work with IchoTolot and Simon.

If you clone a miner during the first frames after removing terrain, the cloned lemming moves into terrain. This causes problems if the underlying slope is very thin, as the cloned miner will just move through the slope and revert to a faller, without removing any terrain. This way the cloned lemming gets separated from the crowd.
See the attached replay for an example of the behavior.

There are a few possibilities to fix this, unfortunately none of them is perfect:
A) Naive cloning of latest miner mask: Just add the terrain removal mask at the position of the cloned lemming, if he has not yet finished moving. The problem appears when the lemming is in the middle of moving down: Then the mask is farther down than it should be and the cloned lemming will not move to the end of the tunnel.
B) Intelligent cloning of miner mask: If the lemming is still at position (0, 0), then just copy the mask at this position. If the lemming is at his final position (4, 2), then don't apply the terrain removal mask at all. If the lemming is at the intermediate position (2, 1), then add the terrain removal mask at position (4, 0), i.e. where the cloned lemming would have started his own cycle. Probably only the lower half of the terrain removal mask has to be used at position (4, 0) (i.e. when the lemming is at position (2, 1)), to avoid removing terrain from the ceiling of the miner tunnel.
C) Naive cloning of latest miner mask and teleporting the cloned lemming: Same as fix A), but if the lemming is at the intermediate position (2, 1), the cloned lemming is teleported to position (-2, 1), i.e. appears at this position instead of the position (2, 1). There are two problems with this: First of all this breaks the rule "cloned lemmings appear at the same position as original lemming". Secondly if the pixel (-1, 2) is non-solid, the cloned lemming should fall down this pixel, but due to the teleporting it will not (at least not automatically).
D) Starting a new cycle for cloned lemming: The cloned lemming starts at the very beginning of the miner cycle, i.e. with removing abother bit of terrain. The main problem is, that this is behavior is not seen for any other skills and allows to go down faster by spamming cloners during the right frames.
E) Stalling the miner: This would be similar to the basher mechanics. One modifies the miner movement, such that the miner stays in his current position in all cases where he would move into solid terrain. The problem here is, that the movement would look pretty weird and the miner actions of original lemming and cloned lemming are no longer symmetrical.

Summary: Fixes A) and C) create additional problems. Fix D) is inconsistent with the other behavior of cloners. Fix B) should (hopefully) not create any problems with the game mechanics, but might be difficult to implement. So fix E) might be the best solution.

Re 1, my proposal was to no longer disallow assignment to jumpers in the first place. ie. if you click a jumper with Builder selected, he immediately starts building just like a walker would.

This is a very bad idea due to the reasons ccexplore (no improvement for downwards slopes) and IchoTolot (more backroutes) mentioned. Moreover allowing to start builders, platformers and especially stackers mid-jump has the potential for lots of incredibly annoying tricks/glitches (I can provide example levels if needed). I would rather keep the old handling than allowing to start skills at some intermediate height.

Perhaps I should have given more background on the reason, why I made this suggestion in the first place:
When already pausing, the frame advance options are already good enough so that further improvements are not really needed. But when I play an easy level where it doesn't really matter where I assign a basher/builder/whatever, I just want to assign the skill. Currently I have to pause the game for this (or hope to be lucky), which I find dissatisfactory. Even if there is no big slope, but only some bumpy terrain (grass in the Dirt style or on rocks in the Rock style) it sometimes happens that a skill doesn't get assigned, due to unlucky timing. When this happens during fast forward or when I press 10sec-skip directly afterwards (without double-checking the assignment), it is very annoying.
So I definitely don't want to add new positions where non-permanent skills can be started, but only ensure that those unlucky skill assignments will still be executed at the earliest moment, when such assignments are regularly allowed.

Assignment to fallers is much more game-breaking than assignment to jumpers...

That is the reason why I suggested to remember the skill assignment for a few frames until the lemming reaches again solid ground, where he can regularly start working.

2. Should the one-assignment-per-frame limitation be kept?

I would put removing this limitation as [LOW-PRIORITY]. While it would be nice to remove the problems with colliding assignments from player and replay, it is not really game-breaking. I cannot think of any (non-artificial) level, where changes here create new backroutes.

It seems only Proxima is in favor of keeping them.

To be fair: In the Gimmick cull thread some more were in favor of keeping secret levels:
- möbius
- GigaLem (though with alternative methods for unlocking them)
- 607 (though with unlocking via talismans and completing ranks)
and perhaps I missed someone else.

I finally managed to fix IchoTolot's backroutes (and a few others) and even brought all my replays up to date . Version 1.1 is attached to the first post.
Levels that were modified: 3-4, 3-13, 3-16, 4-4, 4-7, 4-8, 4-9, 4-10, 4-18, 5-2, 6-3, 6-4, 6-5, 6-20.

For Creatures of the deep in doggycharly's pack, it uses the BeastI VGASpec.

Thanks a lot. Unfortunately that is only half of the problem with this level: I still get an error that some extended graphic set (=VGASpec?) is missing and the level does not load at all. So I cannot even add the VGASpec back in at a later point.

Spoiler (click to show/hide)

is turning a builder with a blocker in such a way that lemming underneath don't walk up the bridge.

  I wouldn't be surprised if this doesn't work in NeoLemmix.

This trick, which does not work in NeoLemmix, the precisely the reason why this level was removed from the NeoLemmix pack. I am fairly sure that RR-fiddling does not allow enough crowd control to solve the level without this trick.

I need to check these out very soon. I have not even heard of until now; - Christóbal Márquez and Alfonso Celis' levels (148 levels) and 68 levels by Anthony Thompson!

I am fairly sure, you already know about the CHRISFN and anthpck level packs, but this shouldn't stop you to check them out.

Anyway, I updated bulletride's levels (removed Silent Circus and Fall Guy) and added the levels by Eymerich/Ellischant, doggycharly, Janne Vieri and Jeffrey Lancaster.

Also I get really sick if I think about the consequences for already existing content (broken levels + creating new backroutes) and replays if I think about what could change here

I agree that we should try to keep the current checks if possible.

However there are a few more things we might consider in order to simplify skill assignments:

1) Instead of having only two lemmings, loop through all lemmings under the cursor
The algorithm would be something like: Get list of all lemmings under cursor (ordered by priority; modified by only-right, etc.). Then go through all lemmings and check whether one can give them the skill, until one has found one such lemming or the list runs out.
The reason for this suggestion is, that currently spamming builders, bashers, ... can only be done with priority-invert. The above would make this easier.

2) Ease skill assignment on steep slopes
Problem: On steep slopes the lemmings are most of the time a jumper or a faller, hence cannot be assigned most non-permanent skills.
2a) Downwards slopes
Here the skill assignment would have to be remembered for two frames (or so). If during this time, the lemming hits terrain again, he would start with the skill.
I am not sure how hard it is to implement such a thing with the current code.
2b) Upwards slopes
The algorithm would be somthing like: If a jumper gets assigned a non-permanent skill, he drops instantly back to the bottom and there starts with the work.
Unfortunately this might have some unwanted consequences:
- The skill assignement can be delayed for two more frames.
- Depending on the precise terrain layout, the drop back to the bottom, might find another different bottom and not the original one. Perhaps one can compute the original position from the frame number of the jumper animation?

I would say, it boils down to this ambiguity: Do we want to help players to recognize steel or do we want to help level designers to create beautiful levels?

1) As mentioned before, level designers can mess things up with both implementations of steel. Only with the current one, players can recognize it better.
2) Lix is a puzzle game, so it should be clear from the beginning, what is steel and what not.
3) We have to distinguish between pieces, that decorate other terrain/steel and decoration pieces that by chance sometimes reach over steel. Moss is of the first kind, because one cannot put is somewhere in the air and expect it to look good. The purple piece is of the second kind: One can easily create decoration in a completely empty area purely by putting such pieces there. So it is decoration, that is a priori independant of the steel, but may reach over some steel pieces.
With the decoration pieces of the first kind (e.g. moss) one might expect that it preserves the original terrain resp. steel. For the second kind (e.g. purple pieces), I would expect that it has one defined behavior, like any other piece that can appear without other context. So it should not matter, whether it is placed over steel or not.
4) I didn't know about that purple steel piece before. Would I have seen it in a Lix level for the first time, I would have expected it to be usual terrain, because I already knew it to be destructible and the rest of Lix works on the basis of wysiwyg.
Sorry, but I feel that adding steel and non-steel versions of one and the same piece is a mistake by graphic designers and only confuses players!
5) I haven't seen much steel decoration in recently made levels. Mostly it was used in the original L1 and in some early Lemmix levels. So I have to ask: Why do we want to decorate steel, when there is so much else one can decorate?
6) When I update old Lemmix levels that have moss over steel, I usually move the moss away from the edge of the steel. This way the decoration is preserved and players only discover the destructible moss by exploding a lemming there (which in many cases isn't terribly likely - Steel Works might be an exception here). And even when the moss gets blown away, the gameplay doesn't change - just one part of the levels looks ugly now.