Methods

Blizzard exposes one achievement per earned seasonal title (“Gladiator: Midnight Season 2”). We classify every achievement in the Player-vs-Player category by name pattern: arena tiers Combatant → Challenger → Rival → Duelist → Elite → Gladiator, with a season-prefixed Gladiator (e.g. “Venomous Gladiator”) marking Rank 1. Solo Shuffle uses Legend (prefixed = R1), Blitz uses Strategist with prefixed Marshal as R1, and rated battlegrounds use “Hero of the Alliance/Horde: <prefix>” — the prefix identifies the season via the matching Gladiator title. A character's title history shows its best tier per season; the notable counters count seasons at Gladiator-or-better, Legend-or-better, and each R1 family.

The milestone dates on profiles are the completion timestamps of Blizzard's rating achievements: “Just the Two of Us” (2v2 1550–2200), “Three's Company” (3v3 1550–2700) and the rated-battleground rank achievements (2000 / 2400). These are earned once, so the date shown is the first time the character ever reached that rating.

Account-wide achievements complete on every character of an account at the same millisecond. We fingerprint each checked character with the (achievement, timestamp) pairs of its most recent 800 account-wide achievements. Two characters in the same region are linked when they share at least 20 fingerprints and at least 40% of the smaller set. Clusters merge transitively. This only sees characters that have been checked on this site, so an account's alt list grows as its characters get looked up. Owners can hide alts after claiming their account; hidden alts disappear from every page, including ladder glyphs and linked-account chips.

For any leaderboard we also compute a rank that counts only the best row per detected account. If three of the top ten are one player, the #11 player is #9 “without alts”. Your own account never counts against you: if your alt sits above you on the same board, the two of you occupy one spot, so you both show that spot's rank. This feeds the unique-spots numbers on the cutoffs page and the multiclass percentiles.

Official cutoffs are Blizzard's own numbers from the season reward API, synced hourly (the page shows when a value last changed). Predicted cutoffs recompute every ladder cycle with Midnight's rules: for Solo Shuffle and Blitz, Rank 1 goes to the top 8 characters per specialization per region among those meeting the activity threshold — the spot count grows as 0.1% of the active pool once a spec is big enough, and when fewer than 8 are active the line sits at the last active character. 3v3 and RBG keep the classic formula: the line is the rating of the floor(0.1% × all rows)-th row (0.5% for RBG) — validated live, where Blizzard's 3v3 value matched that row exactly. The 3v3 achievement's “150 games won” requirement gates claims, not the line. “Spots” counts the rows inside the line; “unique” the distinct accounts among them (alts collapse via fingerprint clusters). A standing invariant fails when the median drift between our predictions and the official values exceeds tolerance; single scopes legitimately wander early-season because official values update on a slower clock.

Per player (alt-group), per bracket family: each spec they field contributes points from its rank-without-alts converted to a percentile of the estimated eligible ladder (unique R1 spots × 1000). Tiers: top 0.1% → 900–1000 · 0.2% → 750–900 · 0.5% → 550–750 · 1% → 300–550 · 2% → 150–300 · 5% → 50–150 · rest → 0–50, linear inside each tier; tank specs weigh ×0.4 (capped 400). The score is the sum over specs. Role tabs recompute over that role's specs only. Click any score for the full per-spec breakdown.

Popularity = players on a spec ÷ players in the rating band; winrate = season wins ÷ games of the characters currently in the band. Bands: all, 2100+, 2400+; periods: season and last-7-days (characters with ladder movement in the window). The tier score compares specs within their role: s = z(popularity) + z(winrate) at the 2400+ band (falls back to a wider band when the sample is small); S ≥ 1 > A ≥ 0 > B ≥ −1 > C. Every table shows its sample size.

Builds aggregate the enriched top-N ladder characters per spec (N selectable 1000 → 25): hero-tree split, the five most common PvP-talent combinations, per-talent pick rates, the most common full loadout with its export code, races, gems, enchants, average secondaries and per-slot item choices. The sample size and the lowest sampled rating are always displayed — small samples mean low confidence, and coverage grows continuously as the enrichment sweep walks the ladder.

Blizzard publishes ladder snapshots, not match logs. We import every board roughly every 30 minutes and diff consecutive snapshots: characters on the same board whose wins/losses changed by the identical amount in the same interval are candidate teammates. When exactly a team-sized group moved together the label is high confidence; larger co-moving groups are labeled possible. Solo brackets never show teammates. Session rows are those same diffs — a burst of games inside one import interval.

The at-a-glance score on profiles and the players database. Arena: current form carries most weight, career peak matters, and the activity term saturates at 300 games so farming games cannot inflate it past its 300-point cap. The RBG pulse is the same formula over RBG + Blitz. Scores recompute after every ladder import.

A guild's score per bracket is the average current rating of its 10 highest-rated members (minimum 3 rated members, otherwise no score). Members are counted on the guild's home realm as reported by member profiles, so connected-realm guilds aggregate correctly.

Ladders: every ~30 minutes per region. Profiles: on demand with a 5-minute refresh cooldown. Achievements & titles: with each profile check. Guild scores: every 6 hours. Alt clusters: every 10 minutes over newly checked characters. Archives freeze at season rollover. The status endpoint reports live job health and data invariants.