I have thus far been unable to find any general winning strategy, even on games containing less than seven squares. For any strategy I can devise, there is at least one game that causes this strategy to fail; a simple strategy to maximize the number of squares of one's own color fails as early as {AABB}, for example (the winning move for Alice is to move to {BBB}, not {B}+{AB}).
Well the winning strategy is probably more along the line of "if you see pattern X, do move Y", so the job of the strategist is to determine what those patterns are, as you would in a game like Othello/Reversi.
Obviously games of less than 3 tiles left are trivial. The game is then deterministic based on who goes first and what the board configuration is. We can assume "A"lice always goes first for the purposes of designing strategies.
AA - Bob wins.
AB - Alice Wins.
BB - Bob wins.
So player 2 always wins 2/3 random games for length 2 boards.
AAA - Alice can go on the left or middle. If she goes on the left, you get BA, then bob wins, otherwise you get B_B and Bob wins.
AAB - alice can go on the left or middle. If she goes on the left, you get BB, and alice wins, otherwise you get B_A and Alice wins.
ABB - Alice has no choices at all. She goes on the left => AB, bob goes on the right => B. Bob wins.
BAB - Alice has no choices at all. She goes in the middle => A_A, Alice Wins
BBB - Bob wins
ABA - Alice can go on the left. Then you get AA, Alice wins.
I've ignored mirror-image games. From what I can tell, none of the 3x1 games actually have a choice that allows the current player to affect winning
or losing. But I can tell you something about the strategy now.
"Islands" are separated 1x1 areas. Since they are either A or B, you want to prevent making 1x1 "islands" for your opponent. Also the 2x1 areas can be exploited, because knowing what color they are and who's turn it is gives you a deterministic evaluation of who will win them. Note, you can ignore any solid-color ones and only have to worry about the AB ones.
So ... AABB with Alice to move (With bob to move is a mirror image of this game, but with B instead of A)
Alice can move AABB -> B_AB. But this creates two separate low-level pieces Bob can exploit. He can flip the AB => B or eliminate the "island B". Notes, that if he flips the "island B", then Alice can flip AB => _A creating her own Island. So when you see an AB and it's your turn you should always flip that before any Islands. Alice shouldn't move AABB -> B_AB because you
don't want to create AB's on your opponent's turn or give them islands of their color. That's your strategy.
Alice can alternately flip AABB -> BBB. But as you'll notice from my evaluation above, when you have 3x1 of your own color and its your turn, your opponent always wins. Alice as an experienced player knows this and creates a BBB that bob needs to move on during his turn, which results in an Alice win.
Let's see if this same logic can be applied to other 4x1 games. Obviously ones with only one "A" are trivial since there's only one choice, they devolve straight to the lesser puzzles. So it needs to be alternate 2A, 3A or 4A puzzles:
ABAB. (result: not a game)
- ABAB => AAB. Rule: Someone (bob) forced to go on the end of a 3x1 loses, so Alice wins.
- ABAB => AA_A. Alice wins this easily, no question.
ABBA (palindrome) => _ABA. Someone forced to go in the middle of a 3x1 wins, so Alice wins. Not a game.
BAAB (palindrome) => A_BB. Bob goes to the BB => A giving Alice two islands so Alice wins. Not a game.
AAAB has more choices (but none of them favor the A player)
=> _BAB. Rule: someone who can pick from left or right of a 3x1 wins. So bob wins
=> B_BB: alice made islands for bob, so he wins.
=> AB_A: better, alice made an island for herself. But still, bob flips AB=>B, alice flips her island, then bob flips his and wins.
Well, I think the strategy for 1xn could just be memorizing the patterns for 1x3 then avoiding making patterns that favor your enemy while trying to make islands that favor yourself. Also, always flip ABs before islands, because ABs favor whoever flips them first.