Reference-reader comparison
Where the community has a mature reader for a format (SoulsFormats and SoulsFormatsNEXT for FromSoftware games, CodeWalker for GTA V, UnityPy for Unity, RRAssetStudio for Riders Republic), the strongest cheap check of a reading is a differential: read every entry with both, and compare the results field by field, or the decoded bytes byte for byte.
What it proves
Section titled “What it proves”A match on every entry proves that two independently written readers agree on the whole install. Combined with a strict census, it proves both the layout and the decoding (for example: every archive entry of Sekiro and Elden Ring decodes to the same bytes as the reference readers produce; see BHD5 and BDT archives).
The reference only proves what it computes. If it reads a field and ignores it, a match says nothing about that field’s meaning.
When the reference is wrong
Section titled “When the reference is wrong”References are not ground truth. A disagreement is investigated against a third authority (the engine’s own code,
the game’s rendering), and when the reference is the one that is wrong, the page says so. Example: older SoulsFormats
reads FLVER vertex member type 0x13 used as a texture coordinate as two 16-bit integers over 2048, but Elden Ring’s
vertex programs read it as four unsigned normalised bytes, unscaled; the programs are the authority.
What counts as independent
Section titled “What counts as independent”- A third-party reader written from its own research: independent.
- Our own earlier tool, a port of the same notes, or a reader that links the code under test: not independent. Its agreement is a regression check, not evidence.
- A tool whose licence forbids copying is still a valid oracle to run and compare against; copy none of its code.
Without a match
Section titled “Without a match”What a reference reader says and nobody has compared is Sourced, with the tool named in the badge.