SimulationCraft WASM
Stat Weights, Stat Priority and Target Error, Explained
Three terms every sim result throws at you, and most guides never define. Here's what they mean, where the numbers come from, and the mistakes people make with them.
What is a stat weight?
A stat weight is the answer to a simple question: if my character gained one more point of
this stat, how much DPS would that add? SimulationCraft computes it by perturbing each stat — simming your
character, then simming again with a chunk of extra crit, haste, mastery, or versatility — and measuring the
slope. The output looks like Haste 1.32, Crit 1.18, Mastery 1.05, Vers 0.98:
relative values, usually normalized so your primary stat is 1.0.
Ranking the weights gives your stat priority — the "Haste > Crit > Mastery" line every spec guide leads with. The crucial part most players miss: weights are a property of your character right now, not of your spec. Stack enough haste and its weight falls (diminishing returns and rating caps do their work); swap a trinket or a tier bonus and the whole ordering can flip.
Weights vs simming the actual item
Stat weights were historically used to score items ("multiply each stat by its weight, add up"). That works for small comparisons, but it's a linear approximation of a non-linear game — it drifts exactly when decisions get interesting: procs, set bonuses, on-use trinkets, large ilvl jumps. Modern tooling skips the approximation and sims the actual item swap: that's what the Gear Compare and Droptimizer modes do, and why they beat any weight-based scoresheet (including in-game addons fed with stale weights).
What is target error?
A simulation is a statistical experiment: run the same fight thousands of times with random procs, crits and timing, and average the result. Target error is the precision knob — it tells the engine to keep running iterations until the DPS estimate's statistical error falls below that percentage. The default here is 0.2%; bulk comparisons in Charts mode lock it to 0.05% for publication-grade stability.
- Lower target error = more iterations = slower but more precise. 0.1% needs roughly four times the iterations of 0.2%, not double — precision costs quadratically.
- Any DPS difference smaller than the combined error is noise. Two items 0.15% apart at 0.2% target error are statistically tied — pick by other criteria (survivability, sockets, aesthetics, whatever).
- Match precision to the decision. A quick "roughly what's my DPS?" is fine at 0.5–1%. Choosing between two close trinkets deserves 0.1% or lower.
This is also what Raidbots meters on its free tier — server iterations cost them money. Running on your own CPU, you can set the error as low as your patience allows; the Smart Sim option keeps big comparisons fast by eliminating obvious losers at coarse precision first, then refining only the contenders.
Keep reading
- Free Droptimizer — apply this to your actual loot table.
- Sim for Mythic+ — weights shift hard between single-target and dungeon profiles.
- Great Vault picks — sim your weekly choices instead of guessing.