How parallel worlds differ

Every seed has three measured worlds — the West parallel world, the Main world, and the East parallel world. This page asks a single question of each of the eleven per-world loot-and-feature axes: does which world you are in change how much you find? It answers with a one-way ANOVA across the three worlds and Welch t-tests between each pair, over all seeds.

Read the effect size, not the p-value. Each world holds ~2.1 billion measured values, so the standard error is minuscule and any difference in means — even a thousandth of an item — registers as “significant” (p ≈ 0). The honest measure is how large the difference is next to the seed-to-seed spread: ω² / η² for the three-world comparison, Cohen’s d for a pair. Those lead every result.

Only the eleven loot-and-feature axes are stored per world, so only they can be compared this way. Spells, damage types and creatures are whole-seed counts with no per-world split — they cannot appear here.

Do worlds differ at all? — every axis at a glance

Each row is one loot axis. ω² is how much of that axis’s per-value variance is explained by which world you are in; max |d| is the single sharpest world-to-world gap in pooled standard deviations. Sorted by ω², largest first.

One axis in detail

Mean per world

Each bar is the mean count in that world; the whisker spans ±1 standard deviation.

Pairwise world contrasts

Every world against every other for this axis: the mean difference, its Cohen’s d, and the (formal) p-value. d is the honest magnitude — how far apart the worlds are relative to their spread.

Every statistic is a full census over seeds — no sampling. Per-world summaries come from each axis's exact population histogram (mean and standard deviation baked in maxStats.json). Only the eleven loot-and-feature axes are stored per world. Effect sizes follow Cohen (1988): d 0.2 / 0.5 / 0.8, ω² 0.01 / 0.06 / 0.14 for small / medium / large.