LMCF LSE is much more complicated than Roux LSE and one of the reason is exactly as people mentioned, that you have cases where the L/R colors are misaligned; doing L2 or R2 means that you can still technically do Roux LSE but if you do a quarter turn on L or R the color recognition becomes basically impossible; this is why LMCF LSE works very differently than Roux LSE so that it doesn't depend on the L-R colors being equal or opposite. In LMCF if you have the case where the remaining unsolved edges are the M-slice plus UL+UR, then you must convert the cube to a case where one of UL or UR contains either of the UL or UR edges in any permutation or orientation, then you solve UL+UR+orient midges in one step.
In 'bad' LMCF LSE cases, both UL and UR contain edges from the M-Slice. In this case it takes typically a 3-move U-M-U style combo to push a random UL/UR edge from the M-slice into any of their UL/UR slots in any orientation/permutation, then use one of the LMCF algorithms to finish. In this fashion the system is invariant of the L/R color alignment. Of course LMCF LSE also allows for the even more weird situation where the unsolved edges are UR+FR+Midges or UL+FL+Midges.
In 'bad' LMCF LSE cases, both UL and UR contain edges from the M-Slice. In this case it takes typically a 3-move U-M-U style combo to push a random UL/UR edge from the M-slice into any of their UL/UR slots in any orientation/permutation, then use one of the LMCF algorithms to finish. In this fashion the system is invariant of the L/R color alignment. Of course LMCF LSE also allows for the even more weird situation where the unsolved edges are UR+FR+Midges or UL+FL+Midges.