How A Bug In Arena’s Rank Calculation Is Making Ladder Exploitable

Back in 2021, Arena player and Twitter user @Hareeb_AlSaq made waves in the magic community with their frequent experimentation on ladder. Most notably, Hareeb would concede an inordinate amount of games in an attempt to understand how the matchmaking system works as it’s not information that is publicly available to players.

One MTGA Player Conceded Hundreds of Matches to Probe the Ranking System – Here are the Results.

Through these experiments, he learned of some exploits that could be used on ladder to get easier pairings and thus easier wins.

Drawing and Conceding Matches Can Abuse the MTG Arena Ranking System – And It’s Causing Problems in Mythic

Then for awhile, it seemed Hareeb went quiet as, perhaps, their experimentation finished. However, just a few days ago, Hareeb came out with another article purporting that they made a big discovery on how matchmaking works and a bug that’s making ladder exploitable.

Why Draw Attention To This Issue?

We spoke about this before in the second article we published about Hareeb’s findings, but this case is slightly different.

Our goal isn’t to help propagate the intentional misuse of a bug in Arena ladder to help players achieve more wins. We are attempting instead to raise visibility on this issue so that it may reach the right people so it may be addressed, then rectified. As Paul initially stated, ignoring the issue won’t make it go away and this issue has already been exploited by many before the publishing of this article. Now, with a greater understanding of where the issue stems, it may be easier for those who can fix it to solve. While I don’t personally purport that I have an extremely strong grasp on the mechanisms Hareeb used to deduce this information, one of their conclusions is that it seems they may not even know that the bug exists, as from a matchmaking standpoint, would not make any sense to implement. If this is true, even if Wizards was aware of the bug, they may not have had the means to fix it if they couldn’t figure out exactly where the issue lied. Ideally, this can help be the catalyst to positive change.

The Five Rankings of Arena

Through information found in the logs of Arena, Hareeb has deduced that there are 5 distinct rankings that each player has (these are unofficial names): Mythic Constructed, Mythic Limited, Serious Constructed, Serious Limited, and Play.

Any game played in ranked or in a pay in event affect the corresponding Serious value and do not change from month to month (in the sense that there’s a time component to whether the rank changes or not, not that it can’t change in that time). Any games played in the Play and Brawl queues affect the Play rating and also don’t change from month to month. These utilize a classic Glicko-2 rating system where new players start at 1500 elo.

Mythic rankings, on the other hand, only appear once you are in Mythic and are only affected by games withing Mythic. Furthermore, it seems your initial Mythic placement is a function of your Serious ranking placement, so by having a high Serious ranking, you’ll enter Mythic at a much higher number. Hareeb was able to deduce this by conceding a large amount of games once hitting Mythic which substantially lowered their Serious and Mythic ratings, then performing well enough in constructed events to help recover their rating to around 1650 (again, the new player start is 1500). Then when they hit Mythic the next month, Hareeb entered at a decent ranking proving that Mythic ratings aren’t conserved between months.

The Bug: Games Have Positive Sum Elo Expectations

As Hareeb points out, Glicko-2 isn’t a zero sum rating system. Matches are played with this system all the time where players gain more elo than the opponent lost and vice versa, however, that’s not what’s happening. Apparently, when you lose a game of Magic, the system correctly registers it that you lose to your opponent. However when you win, the system registers that you beat a copy of yourself rather than the opponent. What ends up happening in this registration of ranking is that games are now positive elo sum. Again, Glicko-2 isn’t zero sum, but it’s certainly not designed to be positive sum. Since the better player in a match is more likely to win, what happens is that they win substantially more points than they should as they’re beating copies of themselves (identical elo) rather than their opponents. If the “weaker” player happens to win, that means they didn’t gain as many points as they should, as again, they didn’t “beat” their opponent, but a copy of themselves This means that if you sustain a high enough win rate, your elo will just keep increasing without an inherent bound!*

*A Reddit user pointed out that they believe they maxed out their ratings at 10k, which as we’ll see in a moment, should not even remotely happen

Hareeb continues that their play rating is 5032, and considering the median is supposed to be 1500, that means they should only lose to the average player, roughly, 1 in 100000000 times. To get an idea of where the elo values should normally lie, you can look at the leaderboard of mtgeloproject.net. As of writing, Allen Wu is the elo leader at 2263.

So What Does This Affect?

So with all this in mind, what happens is what players have known for awhile: if you concede a large amount of games, you’ll face weak players, but we didn’t know why. Considering strong players have obscenely high elos as a result of this bug, the player disparity is magnified greatly. Rather than your average opponent being around 1200-1800, now that is reserved solely for new players or players with abysmal ratings (intentional or otherwise). This is what allows players to tank their ratings to get extremely easy matchups. Hypothetically, based on Hareeb’s findings, one could concede a large amount of matches at a rank floor, cruise up right before Mythic on the back of new players, and then play in constructed events to reestablish a good rating (once again against likely weaker players), and enter Mythic at a solid ranking. This would be extremely time consuming, but not completely unfeasible for someone to do.

Never the less, this is an exploit that allows stronger players to curtail facing equally skilled opponents by artificially decreasing their elo to scores where they will only face completely new players or other people utilizing this exploit. This could also explain how Hareeb has been able to claim the number one spot on Mythic multiple times as, if they had two accounts with low enough elos, they could keep matching up in queue and have one account concede to the other constantly. By extension, since the Mythic rating is “independent” of the Serious rating (the initial Mythic rating is tied in, but after that, are independent values), if they can continue to match with the burner account as they still have relatively comparable elos, this may allow them to cement their placement at the top of the ladder.

Hopefully, now that the exact issue has been discovered, this can be resolved to create a completely fair ladder for players to compete in, especially since qualifications tie into placements.

If you want to read the full report and breakdown by Hareeb, you can do so here.

Thank you for reading!