Match of the Week

The match I selected this time is the underdog match from the latest “Leaderboard Matches”.

Antagonists are Track-3.2, which has been the Top 1 algo for roughly three days and is currently at 2330 elo, and pls-halp2.3, which only has 1474 elo.

It is, again, an epic fail. Track-3.2 always waits out the first turn to adapt to the opponent’s firewall placements.
In a successful match this will detect a corner gun and counter that by building a more efficient corner gun.

This can easily be abused as a vulnerability by an ingenious algo faking to be a corner gun in turn 0 and building out a different strategy afterwards… or just… by pls-halp2.3.
Also, it seems quite confident to me to believe that one’s corner gun will do better than the opposing one :joy:, but maybe that is just me.

Track “detected” a corner gun in this match because it saw the diagonal pattern on a single side, but it actually falsely detected it.

If @RuberCuber’s algo was really smart, it could have actually known that it was not facing a corner gun because information units can never pass the marked location.

After our underdog completely closes up its left side, it is able to destroy Track's pings by just sending scramblers. Fortunately they always meet, which renders the pings useless and enables the scramblers to deal massive damage because they focus the unprotected spawn point of the pings.

Looking at the history of all matches that use this mechanism, I would actually not call this a flaw. The detection algorithm could be improved on, but this is a very rare exception and as long as it does not happen in a competition absolutely negligible.


Haha, this match was really a big oof for me. My previous ping cannon detection system searched for only if the filter line was in place, and then classified it as a cannon. I’ve already fixed this in some of my later algos, adding in other detection systems like if the opponent is spawning grouped pings, whether the pings would travel down the line etc. But this match was definitely a large slap in the face for me, it can show how adaptive algos can be exploited given the right conditions XD.


It is nice to see that your algo is so effective because a large degree of adaptability though. I only see a maze in matches against your algo and it is always a very confident maze (no protection against a lot of strategies). I was wondering how this strat wouldn’t lose to a bunch of other methods, but now I see that the maze is only so clean if it detects that it can be. Nice! It shows how strong a large degree of adaptability can be to provide targeted counters.

[edit] I just looked at my current top algo… and more than 50% of current losses are due to track_x.x algos… :see_no_evil: