TI is such a fun metric. You spend hours throwing arms down and making them do silly dances that seem to accomplish nothing, then suddenly there's a completed solve in front of you. At no point before this did it actually look like it would accomplish the puzzle. 

My immediate intuition for this puzzle, and the basis for this solution, was to pull apart half the molecule, then reset the other half to the start and pull it apart as well. I'd hoped this could be accomplished with 3 arms, but the more I worked on it there didn't seem to be an arrangement that worked for this. Instead, just split the molecule down the middle at the beginning, with the most cringe single-use debonders ever put into a cost secondary solve, and suddenly arm 3 can both move the second half of the snowflake to arm 2, and output the water!

As for further optimization, I'm expecting that a solution with only 7-8 instructions exists using 2 arms, but with 1 day remaining I've yet to find one. the best I've got it a couple designs for 7 instructions that would only work with arms that aren't allowed in game, such as knight's move or length 4 arms, or that, but as of yet I can't find anything that both functions and is actually allowed to exist. There's also a non-zero chance that someone comes around with a 6-instruction solution, but as far as I can find, any such solution wouldn't be able to rotate the input 3 times before resetting its position, meaning that any sensible design would need to somehow deconstruct different pieces each time and yet still output everything, which is outside the scope of things I'm willing to spend my time looking for. If that does happen, congrats on black magic.

I've also made a bet with Morraconda about this metric, so if her solve already showed up, haha I win. if not, I hereby grant morra permission to make fun of me publicly for being bad at TI