It's fully random, so not guaranteed to be solvable.
However, I believe every starting condition should be solvable. ;)
Viewing post in BOXit jam comments
(Slight repost) I made a Boxit solver in JavaScript, and couldn't manage to find an unsolvable starting board after going through something like 10000 starting boards. The full space is 25!, which even when accounting for color exchanges (5!), symbol exchanges (5!), row exchanges (5!), column exchanges (5!), and transpositions (2) is still around 1.6e22, so my search has not been exhaustive by any means.
So, I would be pretty interested if there is, in fact, an unsolvable board :).
Edit: I have constructed an unsolvable board by hand ("fodwqbutimrgkexncvpjyshla"). Still, I wonder if there are other unsolvable boards apart from its reflections. Even better if they are unsolvable because it is impossible to get them down to one tile, not because it is impossible to move from the very beginning. (If this is the only group of unsolvable boards, then landing on it (1 in 1.6e22) would be equivalent to getting ~73 coin flips landing heads.)
Hatnix mentioned it today, very cool! I didn't come around to trying it yet.
I did construct a possibly unsolvable one by slightly modifying the unsolvable one (we had something similar to yours: https://ibb.co/6XDt3bx). Essentially what I did was exchange the two ones in the corners. We manually got it down to two tokens but not solve it yet.
Unless I messed up the board input, swapping the two ones in the corners makes that board solvable: https://ibb.co/wKTcHdp :)