Got past W3 and W2 with an AI that was placing bombs that was calculating a way to clear the most boxes possible within the next 8 turns, did a short work of the bosses.
W1 original boss contained most of us for a while. Got past without realizing when it was promoted to bronze, that same boss became a dummy because it was blowing itself up with its own bombs for a while, so I got inside silver league by not dropping any bomb.
Silver league during most of the week because I have a job and it's taking a lot of my time. I simply did the usual thing I do for almost every contest: Monte Carlo approach. Magus helped me a bit with the patience thing for the heuristic (score += turn_score * (patience ^ depth), 0 < patience < 1) and when properly set up, went through the entire silver league. It was really effective to rack up points much faster than everyone in here.
For most parts until now many others were way ahead of me. I do remember some players asking me when I was finally getting promoted and what was taking so long
Gold boss came up, started working again late Friday. Improved my heuristic a bit, which helped me nearing the promotion. At some point many good players were submitting at the same time, farming on the boss, nearly getting me promoted. But it wasn't enough, and by the time I could start working more, Boss reclaimed a solid 1st 2 points ahead of the 2nd. My AI needed something new, mostly to avoid dumb deaths (such as going in a narrow gap after dropping a bomb with a player ready to block the other side.) I started tweaking my MC to try to predict what others would do (Protip: don't do that) - but it was a failure, most of the time my IA was giving up thinking it was dead when it really wasn't. What I needed was to block risky moves, not calculate everything based on the assumption the opponent would do the said moves.
That's when I came up with the idea to create another quick and short monte carlo search before starting the main one. Listing the 10 possible moves (4 directions + standing still, without then with a bomb) and figure if it could find a way to sneak around and survive the next turns assuming the opponent dropped a bomb. Many times this blocked a single move, helping my bot to avoid obvious traps, and in the worse possible case where no survival possibility was found : ignore that calculation completely and just play ignoring others.
That did amazingly well, I woke up Saturday in the top 30 inside legend. Added a few extra fixes on Sunday that helped me to stay in here, sometimes peeking inside the top 10. I ended up surprising quite a few acquaintances, because I usually never rank that high in a multiplayer leaderboard. Could probably have added a 3rd search to trap others, but I didn't want to risk it.
Overall I really enjoyed this contest, as always with the league system, it really helped me setting progressive goals. I wish I has a bit more time to work on it!