First batch of the benilla-formats goldens, all of them the realm's tables being bigger or retuned
rather than anything misread. The safety check came first and is worth recording: **every
`*_FIELDS` constant in the crate still matches its live DBC header** — Spell 173, AreaTable 25,
CreatureFamily 18, Talent 21, Lock 33 and the rest — so no decoder is reading a moved layout, which
is the only way one of these counts could have been a defect rather than data. Added
`examples/dbc_shapes` to read that off the chain in one go.
Counts: `AnimationData` 208 → 227, `AreaPOI` 339 → 483, `AreaTrigger` 432 → 530 (172/144 across the
two continents, from 133/121), `AreaTable` 1081 → 1481, `GameTips` 74 → 95, cinematics 10 → 13
sequences and cameras both.
Three that are more than a count:
- **Rested experience is ×1.5 here, not ×2.** `Exhaustion.dbc` row 1 is where the client reads the
multiplier it shows, and the realm retuned it. Every vanilla guide quotes 2.0.
- **The help window offers two ticket categories, not ten** — "report someone's behavior" and "an
issue with my character" replace 1.12's Stuck/Guild/Item/… list.
- **Warlock minion pet icons are blank.** `CreatureFamily` rows 15-19, 23, 28, 29 carry no icon
where 1.12 gave each one a path, so `GetPetIcon()` answers nil for a minion and an addon that
concatenates it unguarded gets nothing. Not a column slip: every hunter family in the same column
answers a real path. The realm also backported the TBC families (Dragonhawk, Ravager, Warp
Stalker, Sporebat, Nether Ray, Serpent), added four of its own (Fox, Plagued, Horse, Moth) and
filled two minion ids, Incubus 18 and Felguard 29 — 35 families where 5875 had 23.
And `Emotes.dbc`'s gesture scan, which was pinned on a premise that has stopped holding: 1.12 had
one row per gesture bit, so the walk order could not matter. The realm's file has three rows
carrying `0x8` and a row (424) carrying `0x8|0x10|0x20`, so it matters now — and **our scan is
already right**, which the disassembly settles rather than luck. `0x603a60` walks the table
backwards (`imul $0x1c` to the end, `sub $0x1c` per step) and overwrites each slot, so the earliest
record survives; and its bit test is an if/else-if chain, so one row fills at most the slot of its
first set bit. Ours walks forward keeping the first, with `position` for the chain — the same table.
The test now asserts that rule, re-derived from the file, instead of counting carriers.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SgstFWtn1RxEJ5kVrpgYwn
gromandClaude Opus 5
fbf01c486e