Againrom knowledge
Research-derived documentation of Rage of Mages 1 and selected Rage of
Mages II file formats, interfaces and engine behaviour, with a small claim-reading
tool. The project studies compatibility with independently obtained installations;
this repository is not a distribution of either game or a decoder SDK.
Purpose and method
The findings exist to make an independently written program interoperate with
the original games' own files and runtime contracts: reading their maps,
graphics, texts and saves from a lawfully owned installation, and writing saves
the original game reads back. Againrom's engine is that program; it requires the
owner's installation and distributes no part of it.
Findings come from two methods: observation from the outside (bytes of
installed files, saves written and read back, the running game's visible
results) and static analysis of the original executables, used to explain how
the game reads, writes and acts on those files. METHODS.md
labels every claim as observation, static analysis or mixed, by a
conservative rule stated there: a claim that cites any executable location
counts as static analysis.
The public edition carries no executable locators. Routine names and code or
data addresses are replaced by opaque aliases such as R0001 and L00001;
the mapping to the executables stays in the private research source. An alias
only links statements about the same routine or location to each other.
Sources and publication boundary
The research source policy permits the owner's game installations, their runtime
observations, project probes, and documented general knowledge of standard formats
and platform interfaces. It excludes other game reimplementations and their derived
specifications as factual authorities. The exact research snapshot is recorded in
SOURCE.md.
That is a source policy and provenance trail, not proof that no contributor has
previously encountered outside material, and not legal clearance for every finding.
Hashes establish the exported bytes; reproducing a result does not by itself establish
its historical independence. A separate repository does not establish personnel
separation or retroactively change how an earlier result was obtained.
The functional edition states operations instead of literal instruction text and
omits reconstructable shipped content tables. It does not certify the corpus
as free of original expression. See the change record,
publication policy and component/source distinctions.
Contents
claims/: permanent IDs, findings, confidence, status and evidence references.
formats/: implementation-facing descriptions derived from claims. Scope and
qualifications remain part of each description; a fact about the observed files is
not necessarily a format-wide rule.
tools/claim/: a reader that prints individual claims and their retraction entries.
SOURCE.md: upstream snapshot identity and the scope of this public edition.
scripts/: project-authored, non-mutating publication-review tooling.
The ROM2 functional edition includes all 28 R2-ASSET claim rows and the seven
bounded survey pages. Their claim dependencies are present; their full evidence
remains in the private source named by SOURCE.md. Layout/header
coverage does not establish every field meaning, runtime consumer or protocol.
Reading and checking
Go 1.21 or newer is needed for the reader; it has no third-party module dependencies.
Run from the repository root:
go run ./tools/claim SAV-SACKENTRY-590
go run ./tools/claim -k 'sight range'
go run ./tools/claim -stats
The supplied Python 3.9+ check is heuristic. It reads tracked working-tree files and
reports binary/data candidates, instruction excerpts and unresolved inline links or
claim references, without changing files or printing their contents:
python3 -m unittest discover -s scripts -p 'test_*.py'
python3 scripts/publication_check.py --json
python3 scripts/publication_check.py --strict
--strict also fails on review candidates. Neither a zero exit status nor absence of
matches means legal clearance, absence of embedded game content, or independent origin.
The existing corpus is expected to produce findings; there is no accepted-all baseline.
Confidence and corrections
Confidence concerns the strength and scope of research evidence, not permission to
publish or reuse material. A row may have High confidence for one clause and Medium or
Unknown for another. Read its current wording and qualification, not just the first
rating word.
A claim ID is permanent. claims/retracted.md records withdrawn or corrected claims,
including clause-limited corrections. Publication editing must preserve IDs, confidence
levels, uncertainty and the meaning and scope of corrections. Original evidence and
unabridged research records remain in the private research snapshot.
An experiments/... citation names a private research record at the revision identified
in SOURCE.md; it is not a working public evidence link. Public verification of those
records is therefore limited.
Corrections to facts about the original still originate in the research process.
Publication wording and tooling may be proposed in a public branch. Accepted editorial
transformations must be incorporated into the export process before the next snapshot,
so a raw copy does not silently restore removed material. See CONTRIBUTING.md.
Licensing
The project's own documentation is offered under CC BY 4.0. Project-authored
code in tools/ and scripts/ is offered under Apache-2.0.
These grants cover only rights the contributors are entitled to license. They do not
relicense protected game or third-party expression quoted in a finding, grant rights to
original assets or binaries, or grant trademark rights. This clarification does not
withdraw or add restrictions to the licenses already granted for project-owned work.
See NOTICE and THIRD_PARTY.md.