The Mythic+ meta.

A like-for-like fit per role and metric

Pulls are grouped when they fought the same enemies in the same dungeon, key, affixes and bloodlust state. Within a group, every player is compared with every player of the same role in the group’s other pulls. For every role, one least-squares fit over all of those comparisons estimates a rating per spec so that the difference between two ratings best explains the observed differences. Specs that happened to meet stronger or weaker players are placed on the same scale, and ratings are relative to the role’s field.

DPS, damage taken and healing compare per-second values as ratios on a logarithmic scale. A pull more than 3× above or below what its group and spec predict is clipped to 3×. In a group of N players, each comparison counts 1/N, so every pull counts about once however many pulls share its group. A small ridge penalty pulls thinly covered specs toward the field. Comparisons with the same spec remain evidence but do not inform the fit.

The number on each spec is its fitted per-second value on a typical pull in the selected filters. The typical pull is what the role’s players recorded across these pulls with each player’s own spec rating taken out, and a spec’s rating carries it to that spec, so two specs’ numbers keep exactly the ratio the fit measured. A spec’s raw average would not: it mostly reflects which keys and pulls the spec is played in. Narrower filters, such as a higher key range, move every number together.

The 95% interval comes from the rating’s standard error with the pulls of one run counted together: they share their five players, so they are not independent evidence. A placement whose interval crosses a tier boundary is identified in its evidence view. These ranges describe the fitted values; they do not change the tier’s point estimate.

Tiers describe how far a spec stands above or below the middle of its role on this metric. Every eligible spec counts once. B covers the ordinary range, A and C are above and below it, and S and D mark substantially stronger and weaker results. The bands widen when the role’s results are more spread out. No tier has a quota, and a closely matched field can sit entirely in B.

The centre is the median fitted rating. Spread is the median absolute distance from it, multiplied by 1.4826. The inner boundary is one third of that spread; the outer boundary is the full spread. Minimum boundaries are 1.03 and 1.10× ratios to the median, with reciprocal boundaries below it. Ratios are evaluated on the logarithmic scale. For damage taken and deaths, lower values place higher. Specs without enough comparable evidence do not move these boundaries.

Tier placement requires at least 5 player pulls across 3 source runs and 2 other specs with positive DPS on both sides. These are minimum coverage rules, not confidence guarantees. Limited, missing or unconnected evidence stays unranked.

This describes the available matched outcomes; it is not a causal or buff-neutral estimate. Buffs, cooldown use, party composition and support effects remain part of the recorded values. Healing depends on incoming damage. Each metric is ranked on its own; no combined score is implied.

A duplicate upload of a run counts once. Source runs use the canonical run identity. Dates include both days in UTC, and a pull counts when its run started inside them.