Methodology: where the numbers come from

wow-stats publishes reviews of World of Warcraft PvP seasons once they are over: numbers frozen at the end of the season, always computed over groups of characters or matches. No player, character, realm or guild name appears, and no individual ranking.

wow-mate follows the current season; wow-stats only publishes its review once the season is over.

The ladder: Blizzard's official ratings, collected for rated characters, at the final snapshot on Aug 11, 2026. They feed the rating, rank tier and game mode pages, and the Solo Shuffle and Blitz ladder specs.

Matches: arena combat logs, uploaded voluntarily by players, from Jul 18 to Aug 11, 2026, the last 24 days of the season. They feed the specs, comps and MMR pages.

These matches are not a random sample: most of them come from the top of the ladder. In 3v3, 99% of the opponents faced were met in matches at an average MMR of 1800 or higher.

They describe a finished season; they don't predict the next one. A high win rate in uploaded matches doesn't prove a spec is stronger: it also depends on who plays it and who uploads their logs. When the margins of error of two specs overlap, nothing says which one wins more.

The ladder counts characters: a player with several rated characters counts several times. In Solo Shuffle and Blitz, rating is per spec: a character rated on two specs counts for each.

Ratings of 0 are left out: they only show up for characters rated in another mode. The ladder covers the characters seen during the season; a character never seen isn't in it.

A number is only published from 400 characters, opponents faced or matches, from at least 10 different sources: 10 different players behind the logs, or 10 rated characters. A median or a cutoff (top 10%, top 1%) is only published when at least 10 characters sit on each side of it. The site never gives the highest or the lowest rating: they would point to a single character.

Below these thresholds, the number is hidden, never rounded or estimated. Pages say how many specs or comps are hidden, without their value.

Shares and win rates from matches come with a margin of error: a 95% confidence interval, the range the true number is very likely to fall in. Matches from the same player look alike: same level, often the same partners. So the margin is worked out by drawing whole players at random, with all their matches: it comes out wider, and more honest, than if each match counted on its own.

Ladder numbers have no margin of error: they count the whole ladder as collected, not a sample.

In an uploaded log, the team of the player who uploads it wins more often than the other one. To keep shares honest, a spec's or a comp's presence is measured on the opposing team. Win rates count both teams, and a win rate is hidden when that edge is bigger than its margin of error.

Two scales, never mixed. For matches, the match's average MMR, before it starts: all, 1800+, 2100+, 2400+. For the ladder, the character's rating, at the game's rank thresholds: all, 1600+, 1800+, 2100+, 2300+. “2100+” means “2100 and up”.

In Solo Shuffle, games, wins and losses are counted in rounds, 6 per match. In the logs, a round counts as a match, and the last round of each match is missing: Solo Shuffle match stats cover the other 5.

Every correction to a published number is dated in the changelog, with the page it affects.