How accurate is the ELO rating?

I started wondering this when I reuploaded an algo that had done terribly previously and then suddenly did well.

Looking at yesterday’s first leader board page (the order is not correct, just look at the elo values), we can see that Cthaeh-v2.7 was ranked at position 5 with 1808 elo.
Now, it is ranked at position 22 with only 1729 elo.

Most of the algos have remained and there have not been any notable additions to the top algos.

Does this mean that the leaderboard is mostly based on luck :grey_question:

1 Like

I think the “early” stages of the ELO rating can be pretty random due to all uploaded algos starting at 1500. Even the ones that crash on turn 0. So even if your first matches aren’t to crashing algos, you may be battling other algos that were boosted thanks to crashing algos. Not to mention there’s a bunch of copies of the starter algo.

I’ve noticed some of my early algos tend to go on hot streaks from time to time, and this probably due to luck as their strategies were not very adaptive, so there’s a certain type of opponents that it just works better against.

I believe matches are only within a certain ELO range so you are playing your algo’s peers each day. I assumed that meant things were more stable at the top, but your example of Cthaeh-v2.7 is interesting.


I actually have been waiting for this to happen. I made the 2.7 over a week ago and I’ve known the whole time that it wasn’t as good as my others, but have been scratching my head as it clung to the top 5 spots.
My theory is that since it was uploaded a while ago and got off to a good start, it had the advantage of being older and more established since the rate at which matches occurred was not adequate for the size of the player base.
However, since yesterday it seems like there are considerably more matches occurring for my algos, and 2.7’s plunge in ELO is actually evidence in favor that the ELO system does eventually reach a balance.