I joined the competition about a month late and didn’t care about the top until I was approaching it. The below is just my memory of the history of Halite, it’s shifting meta, and my personal opinions and commentary on it. I apologize in advance if I get some facts wrong, along with who may have started it.

The Beginning:

@djma was the gold standard. Or rather, diamond standard. He was on top of the leaderboards and everyone was trying to beat him.

The erdman era:

I came in during the era in which @erdman was dominating the leaderboard. He was untouchable. I looked at his games and his movements were so fluid, it seemed impossible to beat.

I didn’t know much about his bot as I wasn’t studying it at the time, but I think he held the top spot for a while with @timfoden briefly unseating him, and then @nmalgauti came and dominated the leaderboards.

The erdman + nmalgauti reign:

There seemed to be a constant back and forth between erdman and nmalgauti if my memory serves. I wish I could comment more on what changes were made and what the bots were like, but again they were gods to me and I was a mere gold bot. I had better things to do than be saddened by what they were achieving.

The global meta:

Here’s where I think things started getting interesting. For the a while, good evaluation and smart tactics was able to dominate. @acouette came in and showed the power of finding the global max. This was a game changer as he demonstrated that the short term loss from not taking the best local cells could be offset by going after the global maximum. @nmalgauti quickly followed suit and the two of them were vying for the top spot. This soon because a necessity in order to reach the top spots on the leaderboard.


I don’t know for a fact if every bot employed this. I don’t know who originated it either, but I think it could have been @erdman. But most combat involved sending in your pieces in as a checkerboard formation. This reduces potential overkill if pieces are in sync but maximizes it if not. This seemed superior to not checkerboarding and soon most bots has implemented this behavior.

In comes @mzotkiew:

A surprising contender. @mzotkiew quickly ran up the ranks and dominated the top leaderboard spot. His approach was different. At the time everyone used a pretty standard buildup multiplier around 5. Mzotkiew seemed to do the same and then eventually swap to something as high as 9-12! He often had incredible strength, build it up and captured cells very efficiently. More importantly, his combat was refined. He often took on people with double his strength and could walk away the victor. He demolished the checkerboard combat pattern and I think was one of the first to use STILL on combat squares to try to get overkill damage from pieces moving. Combat would never be the same.

The Non-Aggression Pact:

Who actually did this first I don’t know. There was a post on it from @seelights, and @cdurbin seems to have implemented some form of it in his bot. It wasn’t until @cdurbin and @ewirkerman both had it and were consistently 3-5 on the leaderboard that people really started taking it seriously. In games that they were in together, they generally performed well.

The Non-Aggression Pact Spreads:

The last day of the competition was a complete mess. Holy cow. Pretty much everyone in the top 10 decided to implement some sort of NAP. With that then came interesting meta decisions. A few bots such as myself, @dexgroves, and @timfoden started to experiment with busting through on the last turns. Some bots chose to break the NAP if it felt strong enough. Some then started busting through a few turns before the last. It really was a shame that the servers were so overloaded on the last day of the competition. I feel that with the NAP being widely adopted, a lot of interesting strategy plays could have been made.

The Finals:

Everything is locked in. Who will go down in the history book as the ultimate Halite competitor? Regardless, I’ve had a TON of fun and am grateful I’ve had the opportunity to talk with such talented people.

results matching ""

    No results matching ""