A lazy-man's method, with unconventional 3-look last-layer:
(2) working corner F2L (not really one step!)
(3) "condition edges" (3 distinct sequences)
(4) permute corners (3 distinct sequences)
(5) orient last layer (?? sequences, but not many)
Step (2) used to be corner-edge pairs (it still is for the last pair), but recognition was too slow for me.
After step (3), the edges are in their correct places, either with the orientations correct or with two adjacent
edges flipped. This cuts down the number of cases for step (5).
I had hoped this method would quickly get me down to 30s, but I'm not that fast, owing to lack of talent and skill. I do practise quite a bit.