Reference sheet
Every target, its setup, and the one rule that catches people out. Buffer DF, target UB.
Every algorithm on this sheet is machine-checked against a cube model, not transcribed on trust: each one is confirmed to exchange the buffer with its target and apply M2's side effect exactly. The source table is the Speedsolving wiki.
Setup the target to UB, do M2, undo the setup.
M2 exchanges DF with UB. Everything else on this sheet is just getting your target
into UB and putting the cube back afterwards.
M2 also swaps UF with DB, and the U/D and F/B centres. That side
effect cancels on every second shot, which is why an even number of targets is needed — and why
an edges-only solve, with the corners already done, can never hit parity.
Learn the four shapes rather than the sixteen strings. Two carriers — U to bring the target up to UB, B to carry it round the back — each with an R and an L version. Within a group only the middle turn changes. No rotations anywhere, so the cube never leaves its orientation.
Six stickers sit on the M slice itself: A and Q (UB), C and I (UF), W and S (DB), plus the buffer's own U and K, which are never targets. These cannot use a setup, because the setup would move the slice they live on.
The pair rule. Count your targets from one. If C or W falls in an even position — the second half of a letter pair — use the other one's algorithm instead. Same for I and S.
This is not a quirk to memorise blindly: by then M2's side effect has swapped UF and DB, so the piece you want really is in the other place.
Measured end to end: with the pair rule, 2018 of 2018 even-parity scrambles solve completely. Without it, roughly one in four.
I, Q and S are the long ones — ten and eleven moves, because no shorter sequence exists: a search over every face and slice turn finds nothing within eight. They are also the rarest, at about 4% of targets each.
U and K are the buffer's own stickers. A shot exchanges DF with a different slot, so shooting DF to itself does nothing.