V4 precook check — round 2

TL;DR: Full dump of task/stats_run_round2 (schema v4) for verifying the record layout and the numbers. Everything on this page is decoded straight from the chunk bytes using only the run's own manifest. The source data lives under the gitignored task/ tree, so this page only has content when built locally with that run present — on the public deploy it shows a placeholder.

Manifest

Dimension orders as stored (index order matters — it is the byte order inside each record):

v3 vs v4 — field layout side by side

v3 is the production schema (src/seedScheme.json, bytes). v4 is this run's manifest ( bytes). Rows are matched by field name; a highlighted cell means that field differs between the two schemas.

Dimension orders side by side

Record layout

Integrity checks

The two spell identities are diagnostics, not assertions — if either fails for many seeds, that may be a real property of the data rather than a layout bug, but a wrong offset would typically break them for nearly all seeds.

Population aggregates — every leaf field

Min/max/mean over the whole run for each stored scalar. A field that is always 0, or has absurd max values, usually means the layout is shifted. “at ceiling” counts values stored at the type maximum (255 / 65 535) — saturated floors.

Seed inspector

axes — per-world POI counts

Per-world scalars + spell identities

hordeCreatures

spellDamageTypes

spellTypes

factionCreatures

Raw bytes — every leaf of this record

Byte-level view of the selected seed’s record: each scalar’s span inside the record, its little-endian hex bytes, and the decoded value.