Rules of the contest

Every build is simulated under the same fixed conditions. Only gear, talents, consumes, spec options and rotation come from the submission.

Encounter

Profile: Phase 1 single target (phase 1, classic_era).
Fight length: 120 s, plus or minus 5 s per iteration.
Single target only. One level 63 "Level 63 boss", 3731 armor, 127,393 health, 805 attack power, 2 s swing, parry haste on.
Execute windows: 20% of the fight under 20% health, 25% under 25%, 35% under 35%.
Tanks in the raid: 0. Target dummies: 0.

No world buffs.
Raid buffs and debuffs are fixed by the profile; the submission's own values are ignored.

Fixed by the profile. Values in the submission are ignored. No world buffs.

None.

Policy: any. Any consumes the engine knows are allowed. When a submission carries none, the spec default set is used.

The challenger and the current champion run with the same seed and the same number of iterations. Iteration i of both runs uses the same random stream, so the two DPS values of an iteration differ only because of the builds. This is common random numbers.
For each iteration the difference challenger minus champion is recorded. The published margin is the mean of these differences; the confidence interval is 1.96 standard errors of that mean, from a paired t test.
Stage 1 runs 50,000 iterations. If the mean difference is larger than 2% of the champion's DPS and the interval excludes zero, the verdict is final. Otherwise stage 2 runs 1,000,000 iterations.
After the final stage: a win when the whole interval is above zero, a loss when it is entirely below zero, a tie otherwise.
A win makes the challenger the champion and records the margin over the previous one. A tie adds the challenger as a co-champion next to the current ones; co-champions are statistically indistinguishable under these rules. A loss changes nothing.
The first accepted build for a spec becomes champion after a solo run.

Published precision: DPS as integers, margins and intervals to one decimal. Nothing finer is reported because nothing finer is supported by the runs.

The class must match the spec, gear must not be empty, every item id must be known to the engine, and the rotation must parse and survive a one-iteration dry run.
A build identical to one already run for this spec, profile and engine version returns the existing submission instead of a new job.
A 1,000-iteration solo run rejects builds at zero DPS or below 25% of the current champion.

vanilla engine: the wowsims classic model as is. Nothing about this spec has been checked against Forever.
unverified: Forever changes for this spec are known and modelled, but not measured in game.
verified: the model for this spec has been checked against in-game measurements.

Every record carries the engine version it ran on. When the engine changes, standing records keep their version and can be re-run.

Single target only. No cleave or multi-target ranking.
No world buffs.
Tanks and healers are listed but not ranked. There is no head-to-head metric for them yet.