World of Warcraft Mythic+ Dungeon Analysis



May 21, 2024


The ZugZugCreation project collects data from raider.io and performs various analyses. It began as an analysis of subcreations M+ site, but has since evolved a bit into its own monster.

It’s written by a Mistweaver Monk1 player who got fed up with being in the “F-tier” all the time, and decided to look into how data was analyzed. From there, things took a turn and it has since diverged into a full analysis of the M+ game mode.

1 If you need to get in touch for some reason, my email address is DeadlyDog@protonmail.com.


By the name of Hades! Olympus, I accept this message!

Season 4 is upon us, and things are still shaky on this site. Expect rough edges until every small part has been fixed.

— Melinoë


Data is fetched throughout the season, but more frequently at season start. The following table is updated whenever I pull in a new data set and run it through the site. Early season I might mark some data sets as preliminary if they don’t contain that much data yet.

Section Date
Leaderboards (>1500) DF, Season 3, 2024-03-07
Leaderboards (>2700) DF, Season 3, 2024-03-09
Dungeons DF, Season 4, 2024-05-21

The “Dungeons” data set is used for Subcreation, Dungeon analysis, and Specializations via Dungeons. The “Leaderboards” are used to compute specialization power from leaderboards. The sampler either samples in the population of M+ players above 1500 rating for a casual data set, and above 2700 to capture players closer at the top.

These data sets cover 3 major cases: A general view of specs, but where the noise at the very bottom is removed. A dataset with the better players in the game, and one at the very top, split by dungeon. The views provided yield slightly different results to interpret.

TL;DR - Summary

Class balance is rather good! Most of the preferences has to do with comfort and community perception2, more than actual strength. In many cases, the data doesn’t show any large difference in specialization power when directly comparing them, and the effect is rather small. Hence, you can conclude that the player skill matter relatively more than the choice of class.

2 The community often decides based on popularity moreso than strength. While strong classes are generally popular, it isn’t always the case, and at times, incredibly strong classes do not see much play.

3 Infinite scaling ensures that sooner or later, certain specializations will eventually run into hard times. The amount strategies for beating the content gets gradually more cheesy, and you start to rely more on breaking certain mechanics rather than playing them as they were intended. Historically, Blizzard has allowed this as part of the M+ game mode as long as it doesn’t completely break a mechanic.

If you are running at key level +10 and below, any specialization is viable3. It is when you go to the limit and look at the highest keys that you’ll see a difference in what is being brought. Even at the top level, you might have several specs being viable, and a given composition is chosen due to emergent effects moreso than individual specialization power.

This site provides some evidence for the above statements.


The following is a list of major changes done to the site. It can be used as a way to quickly drill down and find new or altered stuff.

Dragonflight Season 4

  • 2024-04-28—New theme. This one has a bit more contrast between the background and the text which hopefully improves readability a bit. Also pulling a new set of data from M+. While here, we updated the code base which renders the site to the newest version.
  • 2024-04-26—Updated the site with the first batch of Season 4 data for dungeons. Leaderboards will follow eventually, once they settle down a bit, which typically happens 14-21 days into a season.

Dragonflight Season 3

  • 2023-12-06—Added some text around popularity noise. Popular specializations are prone to the introduction of noise at lower difficulty levels to the point where it messes with the statistics.
  • 2023-11-30—Another edit-pass over the later sections of the site. Lot of smaller tuning on plots, text have been fixed in a lot of places and doesn’t refer back to Dragonflight Season 1 anymore. This breathes some much-needed longevity into the text. A ROPE-plot was added to aid in the discussion about over/under-powered specializations.
  • 2023-11-24—Take a editing-pass over the first few sections of the site, up to Subcreation. This edit pass adds some margin-commentary as well as fixing the main text. In the background, much of the internal structure is switched from RMarkdown-style into Quarto-style.
  • 2023-11-21—Switched the site generator from RMarkdown into Quarto. This allows for more advanced layouts, where the margin gets used more.

Dragonflight Season 2

  • 2023-11-11—Added a small section on the state of the meta for season “2.5.”
  • 2023-07-23—Some wording in the earlier sections have been updated to be more generic and less about DF Season 1. This lets us use the material going forward in time in a better way.
  • 2023-07-21—More work is being done on the compositional analysis. Some internal changes have been done to the way we call stan, which should be more future-proof.
  • 2023-07-11—Final snapshot of 10.1.0. Added some compositional analysis based on Correspondence Analysis methods. Added a section about the State of the M+ meta which is more opinionated.
  • 2023-05-23—Pulled in Season 2 data from leaderboards.
  • 2023-05-22—Week 2 data. The number of runs per dungeon was increased to 1000 for the TopK dungeon runs.
  • 2023-05-13—First dungeon pull from Season 2 added. Added a section with more subjective opinions.

Dragonflight Season 1

  • 2023-05-09—Updated dungeon data for the pre-release week. There might be slight problems in some plots this week due to implementation of season 2.
  • 2023-05-05—Added a section which address some of the critique this site has received. While here, switch to a stan-model which partially pools specs. Ready the site for Season 2.
  • 2023-04-25—Further improvements on the “Specialization via Dungeons” model. Flesh out the text in the chapter as well.
  • 2023-04-14—The “Specialization via Dungeons” model is now a nested model, based on individual players. Some initial text has been written as well.
  • 2023-04-09—First steps in building a stan-model for the same dungeon data as is being used by Subcreation. The model is suboptimal, but it is a start.
  • 2023-04-05—Pass over the Specialization chapter, tuning the text.


We are using a rather permissive license for the work presented here4.

4 As we were building the site, we were using a far more restrictive license, preventing commercial use of any form. Now, with most of the site fleshed out, and having a more foundational ground, the license have been changed to a quite permissive one. You are basically free to use this as long as you give proper attribution from where you got the data.

This work is licensed under a Creative Commons Attribution 4.0 International License