Read the count before the chart

Published v4 seed files supply individual seed counts. The September 27 full v4 report contains 483 metric/world references: 69 definitions across all seven selections of West, Main, and East. Its statistics, per-value lookups, and top/bottom-100 lists are already calculated and independently checked across all 2,147,483,647 archived seeds. See the complete v4 reference.

The existing v4 Records view supplies record holders, typical loot, category balance, and Challenge builder’s shortlists. It uses 270 references for individual worlds and the three-world sum. The fresh scan confirmed its rankings. Page scopes and the complete reference’s expanded scopes are stated separately.

v4 is the dataset. Its numbers do not become stale with time. An export timestamp describes a file’s creation, not the validity period of its counts.

Some reference files still need validation or replacement. The existing histogram file disagrees with the verified v4 frequencies. Joint, correlation, and grid exports also need an explicit v4 verification record before their figures can be treated as v4 results. This is export correctness and integration work; these files are not alternative datasets to choose between. Seed positions require a verified matching reference.

The v4 round-2 snapshot contains seeds 1–40,000 only. Its page reads a checked-in test slice. Its means, minima, and maxima describe that subset; they are not substituted for full-v4 statistics.

• West (−1), Main (0), East (+1) are the three measured worlds.
• Three worlds, or existing labels saying all worlds, mean the sum of these three—not every possible parallel world.
• A three-world sum has its own distribution. Its mean is additive; its median and percentiles are not generally the sums of the separate world medians and percentiles.

Images come from verified noita.wiki.gg pages. Damage images come from the wiki’s damage-type icons. Spell-type images are example spells verified in each type, not empty slot backgrounds or official category icons. Creature sprites are member illustrations chosen from the faction or creature-group lists, not official category icons. Counters without verified artwork remain text-only. Text labels stay visible, and existing in-game translations have not been replaced.

The generator’s named counters—such as highSlotWands and hvSpells—retain their existing names. The manifest does not specify every selection threshold, so the UI does not invent one. Faction and horde counts are separate classifications; do not add both tables together. Spell-type buckets may not cover every recorded spell; the v4 audit exposes that difference instead of forcing a total.

Count: the stored number for a category. Zero is a value; a dash means missing or undefined. Counts of spells or creatures are not counts of distinct species or spell designs.

Mean: add the counts and divide by the number of seeds. Rare high values can pull it above a typical count.

Median / p50: the nearest-rank middle count. p90 is the smallest count at or below which at least 90.00% of seeds fall. Neither is a percentage of the maximum.

Mode: the most common count. Several counts can tie; the complete export keeps all tied modes.

Typical range: percentile bounds describe where counts usually fall. In the completed Main-world hearts reference, the median is 71 and p5–p95 spans 58–84. Because counts are integers and ties are included, the actual share inside the bounds may exceed 90%.

Standard deviation (SD): spread of the counts around their mean. It is not uncertainty in the mean. An SD range is not automatically an interval containing a specified percentage of seeds.

Population position: seeds below a value plus half the seeds tied at that value, divided by all seeds. If 80 of 100 seeds are lower and 20 tie, the midpoint position is 90.00%, even though only 80.00% are strictly lower. The distribution page shows lower, equal, and higher counts separately.

Tail frequency: the share at least as high, or at most as low, as the observed count. Inclusive tails include ties. They are descriptive frequencies, not significance-test p-values.

Correlation ®: linear association between counts. Negative means opposite directions; positive means moving together. Near zero does not exclude a curved relationship. Correlation between a total and one of its parts can arise by definition.

Paired difference: world B’s count minus world A’s count for the same seed. Its spread is not interchangeable with the separate spread of A and B. The world comparison pages use the stored paired differences, rather than pretending the worlds are independent samples.

• Histograms and joint world bins: exact seed tallies, sometimes grouped for display.
• Seed-number heatmaps: normalized, quantized densities. Use them to spot possible structure, not to recover exact cell counts or prove a defect.
• Normal-fit diagnostics: compare observed quantiles with an approximation. Integer steps and small deviations are not automatic failures, and finite percentile probes do not test the most extreme tails.

For byte order, field sizes, and raw world-preserving decoding, see Binary data layout.