In my personal experience, main time loss on longer categories comes from stalling a lot on desicion making process. All of that comes from the lack of information about the board, since generally only 2 main rules are applied in order to fill cells ("break" the 3 consecutive colors and fill the rest of the colors in the row/column), similarity between rows is actually very inconvenient and situational concept to use in a run.

Soooo... what if there's a new way to get correct information about the board? Since the game itself is based purely on math, probably it has some undiscovered invariant (perhaps, a hidden rule would be more understandable), which can reduce the run time significantly.

If we use the "gamba" strategy (just putting random colors in the cells, hoping to solve the board correctly), high 0:3X on 12x12 and mid 0:2X on 10x10 is unlikely, but technically possible because there's no time spent on calculations. We can actually get to those times without having luck involved if a new rule is found.