1/26/2024 0 Comments Mtg proplayer winpercentage![]() Also, if nothing else is going on, like a baseline drift, the rating value of 0 is equivalent over any timescale within the same season. For instance, a minimum value loss that knocks you from 33% to 32% at time T will leave you with a similar rating, within one minimum loss value, as a 33%-32% loss several hours later. Under the assumption that the #1500 rating does not change wildly over a few hours in the middle of the month when there are well over 1500 players, it’s possible to benchmark a rating without seeing it directly. Deranking to absurdly low levels let me run several experiments. Thankfully, this guess turned out to be correct. I hoped that the minimum win/loss would be quantized at a useful level once the rating difference got big enough, and if true, it would allow me to probe the algorithm. I’ve decided to structure this as a journal-of-my-exploration style post, so it’s clear why it was necessary to do what I was doing if I wanted to get the information that WotC has continually failed to provide for years. Scroll down- way down- to the Problems section if you want to skip how I wasted too much of my time. The explanations for what’s actually going on are more technical, and the next couple of sections are going to be a bit dry. The rating system seems to be effectively the same throughout the last couple of weeks of the month, at least in Mythic. As it turns out, neither of those appear to be directly true. Either of these would explain the massive collision of people outside the top-1200, who are playing, into the the people inside the top-1200 who are trying to camp on their rating. calculating the correct rating changes, then giving one or both players a small boost to reward playing). I suspected that WotC might be rigging the rating algorithm as the month progressed, either deliberately increasing variance by raising the K-value of matches or by making each match positive-sum instead of zero-sum (i.e. The actual system does multiple things that are not Glicko and does not do at least one thing that is in Glicko. ![]() Glicko is a modification of the old Elo system. The only two pieces of information we have been given are that 1) Mythic Percentile is the percentage (Int(Your Rating/#1500 rating)) of the actual internal rating of the #1500 player. ![]() Were clones of the top-600 all sitting around waiting until the last day to make Mythic and kick everybody else out? If they were already Mythic and top-1200 talent level, why weren’t they mostly already rated as such? The decay is also much, much worse than it was in late 2019, and those explanations give no real hint as to why. While both true, neither seemed sufficient to me to explain that level of decay. The canonical explanations were that people were grinding to mythic at the end of the month and that people were playing more in the last couple of days. There’s significant decay in the days leading up to the last day, just not at that level of crazy. These days, a top-600 rating with 24 hours left is insufficient to finish top-1200, and it’s not just a last-day effect. It’s well known that the rank decay near the end of the month in Mythic Constructed is incredibly severe. ![]() The change making Gold +2 per win/-1 per loss likely turbocharged the issues from #6.Giving everybody similar initial ratings when they make Mythic leads to issues at the end of the month.Performance before Mythic seems to have only slight impact on where you’re initially placed in Mythic.Early in the month, the system is more complicated.The minimum change in rating is ~5 points in a Bo1 match and ~10 points in a Bo3 match.After convergence, the Bo1 K-value is ~20.5.That comes out to around 11% Mythic later in the month. Expected win% is constrained to a ~25%-~75% range, regardless of rating difference, for both Bo1 and Bo3.There is no detectable Glicko-ness to Mythic constructed ratings in the second half of the month.It’s just as exploitable as you should think it is.There is rating-based pairing in ranked constructed below Mythic (as well as in Mythic).The details are also seriously nerdy, so reading all of this definitely isn’t for everybody. If I get enough data, I can make a rank-decay curve for every rank at once, among other things.īrought to you by the all-time undisputed king of the percent gamersĪpologies for the writing- Some parts I’d written before, some I’m just writing now, but there’s a ton to get out, a couple of necessary experiments weren’t performed or finished yet, and I’m sure I’ll find things I could have explained more clearly. Note: If you’re playing in numbered Mythic Constructed during the rest of May, and/or you’d like to help me crowdsource enough logfiles to get a full picture of the Rank # – Rating relationship during the last week, please visit and DM/share. Edit: A summary, including some updates, is now at
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |