How the numbers are produced, why each choice was made, and where the limits are.
Where the data comes from
Every number on this site comes from runs that players actually logged to Warcraft Logs. Nothing is simulated and nothing is estimated. The scraper pulls newly uploaded Mythic+ reports once per hour and stores each player's damage per dungeon run.
Only timed runs count. A depleted key usually means something went wrong, which distorts damage numbers.
Only key levels +13 and up are stored, grouped into the brackets you see on the site.
One week at a time. A week runs from the EU reset (Wednesday 05:00 UTC) to the next. Past weeks stay available in the week selector.
How a spec's number is calculated
Two values are shown per spec. Both are calculated the same way:
For each dungeon separately, take the median of all that spec's runs.
Average those per-dungeon values into one number.
The solid bar is that median, described on the site as the standard run. The faded bar is the same calculation applied to the top 5% of runs, which shows what the spec reaches when things go well.
Averaging per dungeon rather than pooling every run matters because dungeons are not equally popular. Without that step, a spec that happens to be played a lot in a high-damage dungeon would look stronger than it is.
"Same calculation" is meant literally: the top 5% is taken per dungeon too, then averaged. Every dungeon a spec was played in contributes its own best runs, so the faded bar can never come from a single dungeon. The best run of a dungeon always qualifies, which has a consequence worth knowing: where a spec has only a handful of runs in a dungeon, its "top 5%" there is that single best run. The run count next to each spec, and again inside the per-dungeon breakdown, shows when that is the case.
Key levels inside a bracket are pooled, and the best runs come almost entirely from the highest key in it, because more enemy health means more damage. Read the faded bar as what a spec does on the bracket's top key rather than as the best 5% of the lowest one.
Why median, and why top 5% instead of "top 200 runs"
The median is used because a plain average is dragged around by outliers, and a single broken parse can move it noticeably.
The top 5% is a proportion, not a fixed count, and that distinction is the important one. Specs differ enormously in how often they are played. In one week at +20-22 the run counts ranged from 27 to 19,709. A fixed cutoff such as "the best 200 runs" therefore means something completely different per spec:
Spec
Runs that week
"Top 500" would be
Unholy Death Knight
18,811
its best 3%
Feral Druid
1,116
its best 45%
Outlaw Rogue
583
its best 86%
The popular spec would be compared at its peak while the rare spec is compared at roughly its average. Measured proportionally instead, those three specs land within 1% of each other. A fixed count would have invented a gap that only reflects popularity.
Why healers and tanks are ranked by damage
Healers and tanks appear in a damage ranking on purpose. Their damage is surplus output: it is what they contribute once their actual job is handled. That makes it a reasonable comparison between specs of the same role.
Healing output is available as a separate toggle in the Healers tab, with the caveat below.
Healing (HPS) and why it needs a caveat
Healing per second is not a clean quality measure in Mythic+, because it is driven mostly by how much damage the group takes. A better group takes less damage, so its healer heals less and posts a lower HPS. High healing often means messier runs rather than a better healer.
It is shown because the throughput comparison between healer specs still carries signal across hundreds of runs, and because people asked for it. It is not shown as a quality ranking.
What a spec needs in order to appear
A spec needs at least 10 runs, spread across every dungeon of the season. There is no relaxed alternative. Full coverage is required because each dungeon counts equally in the average, so a spec that never ran the weakest dungeon would score higher for that reason alone. Earlier the site fell back to "5 runs across 3 dungeons" when few specs qualified, which put three-dungeon averages next to eight-dungeon averages and made the ranking measure coverage as much as performance.
The cost is visible right after the weekly reset: on Wednesday morning only a handful of specs qualify, and the list fills up over the following day or two. The top bracket stays shorter all week simply because very high keys are rare. The page says so when that happens rather than quietly lowering the bar.
If a spec is not listed, you can still see it: pick a single dungeon, and every spec with a logged run there appears, since comparability across dungeons is no longer at stake. Clicking it there opens its per-dungeon breakdown, and the dungeons it has not run are listed as "no runs" — which is exactly why it is absent from the overall ranking. Specs in the overall ranking never have such gaps, by definition.
The run count is printed next to every spec, and inside the per-dungeon breakdown for each dungeon, so the sample size behind any number is always visible.
What this site deliberately does not do
No tier list. A ranking with visible run counts lets you judge the data. Compressing it into letter grades hides how thin or solid a number is.
No mitigation or "tankiness" ranking. Damage taken measures where a player stood, not how durable a spec is. Avoiding a mechanic and mitigating it look identical in the logs. This is a question simulations answer better, because they can hold the incoming damage constant. Logs cannot.
No fixed-count filters such as "top 200 runs", for the reason shown above.
Known limitations
Logging bias. Only logged runs exist here, and lower keys are logged far less often than high ones. The data reflects players who log, not all players.
Group dependence. Damage depends on group composition, routes and pull sizes. Buffs from other players cannot be separated out.
Thin early week. Right after the reset the sample is small and the ranking moves around. It settles over the following days.
High keys are sparse. The +22-25 bracket always has far fewer runs than the lower brackets, so treat it as a tendency rather than a precise ordering.
PTR data is a rough outlook. The Season 2 preview covers only the last 7 days and shifts with every balance patch. It is not comparable to live numbers. Key levels +15 and up are pooled into a single ranking, because there are not enough PTR runs to split them into brackets. Pooling key levels has a cost: higher keys mean more enemy health and therefore higher damage, so a wider span measures the key level as much as the spec. The floor sits at +15 to keep that span narrow. Dungeons are not pooled — the preview uses the same per-dungeon calculation as the live ranking, and a spec must have runs in every PTR dungeon to be listed. That requirement is strict on purpose: PTR dungeons differ enormously in damage output (roughly 247k in Altar of Fangs against 161k in Kings' Rest), so a spec that skipped the weakest dungeon would score higher for no reason other than which dungeons it happened to run. Fewer specs are listed as a result.
Balance patches. Numbers within a week can span a tuning change. Weekly snapshots keep those weeks separate but cannot undo the effect.
Season 2
Season 2 starts on 18 August (NA) and 19 August (EU). The site switches to the new season's dungeons at the EU reset. Weekly trend charts return then, because a fresh season gives a history that is consistent from week one. They are currently hidden: earlier changes to the collection method created an artificial drop in the charts that said more about the method than about the game.
Recent changes
Healing (HPS) toggle for healers, live and on the PTR preview
Per-dungeon breakdown when you click a spec
Week selector for the last four weeks
Reset day now shows the current week's data instead of the previous week
Top bracket extended to +22-25
PTR preview now starts at +15 instead of +12, for a narrower and more comparable key range
PTR preview now calculates per dungeon like the live ranking, and supports the per-dungeon breakdown on click
Every ranking now requires runs in all dungeons — the relaxed fallback is gone, so all listed specs are directly comparable
The per-dungeon breakdown now shows which dungeons a spec has not run, and opens in the single-dungeon view too
Specs with few runs now carry a marked run count, so a thin sample is visible at a glance
Shared links keep the full view: key level, role, dungeon, healing toggle and week
Past weeks are now recalculated from the raw runs with the same rules as the current week, and support the dungeon filter and healing view
Fixed: the Magister's Terrace dungeon filter returned wrong numbers
Fixed: mobile layout issues with the share button and long dungeon names