100parse

Notes

How Arms Warrior was built

19 September 2026

Arms Warrior is the eighth spec and the first second spec inside a class. Building one takes three things: a spec file that says what the spec can wear and what each stat is worth to it, a ceiling, and a parse curve. The first two are decisions. The third has to come from somewhere, because a brand new spec has no runs to cut a curve from.

The spec file

Arms started as a copy of Fury's file. Same armour proficiencies (plate, mail, leather; every cloak counts as cloth so the back slot says so). The stat weights are Fury's: strength leads, then crit and hit, hit capped at 9% and worth half past it. Published Arms values sit within a few percent of Fury's on every stat, and the thing that actually separates the two specs in a draft is the weapon.

So the weapon rules are where the file differs. Arms is two-hander only: the one-hand slots are closed and a two-hander fills both hands with one pick, the same construction Ret uses. Mortal Strike deals weapon damage per swing, so at equal DPS the slower two-hander is the better one, and the file leans on weapon speed harder than Fury's does. The ranged slot stays; a warrior still carries a bow.

The ceiling

Every spec has a theoretical best set: the highest-scoring set of raid items the model can assemble, solved once and stored. Its DPS is the ceiling your results page measures a run against. The ceilings are set deliberately so specs sit on rungs rather than wherever the raw maths lands: Fury 1275, then Mage, Rogue and now Arms at 1250, Warlock 1225, Hunter 1175, Shadow 1160, Ret 1150. Arms lands on its rung with one multiplier on every gear term, so nothing about the item comparisons changes, only the scale. Its best set is Might of Menethil with four pieces of the Naxxramas tier set.

The curve, borrowed

A parse is a percentile against real runs on the same spec. Arms had none. The way round it is the same one used for Ret and Shadow: play the new spec and its closest sibling with the same simulated drafters on the same random seed, so the only difference between the two sets of results is the spec itself. That gives a ratio at every point of the distribution. Then take the sibling's curve, which is made of thousands of real human runs, and carry it through the ratio. The result is a curve for Arms that has the shape of Fury's human data with Arms's mechanics applied.

The ratio came out close to flat: Arms sits about two percent under Fury at the same skill, which is what a slow two-hander against a pair of one-handers should do. The bar that falls out of it is 1144.7 DPS, one 100 per 152 runs, a median of 87, and 0.916 of the ceiling, right between Fury's 0.913 and Ret's 0.911. The curve file says "provisional" in it. When Arms has enough runs of its own, its real curve replaces the bridge and nobody's parse moves in the process.

Same puzzle?

The worry with a second spec in a class is that it is the same draft twice. It shares Fury's entire item pool. So before it shipped we compared the two on the same banked daily boards over 29 days: 43% of bosses shared, but only 30% of the winning items. A slow two-hander and two one-handers want different drops, and the rest of the line follows. Seven days in ten the two warriors are solving a different board.

More notes