Technology
How RatOrNot works
A scan takes a few seconds and never runs the file. Here is what happens to a jar after you drop it in.
1. Safe archive reading
The jar is read as an archive in memory with strict limits on size and depth. Malformed archives, misleading file extensions, extra leading bytes and nested jars (including Jar-in-Jar bundles used by mod loaders) are handled without crashing, and nested archives are scanned as well.
2. Mod metadata
The scanner reads the mod loader's metadata (Fabric, Forge, NeoForge or Quilt) to learn which entrypoints the mod declares and how the code is meant to start.
3. Bytecode analysis
Class files are parsed directly. RatOrNot looks at which methods are called, which constants are used and how values flow between them, rather than only searching for suspicious words.
4. Deobfuscation
Common string-hiding patterns are evaluated by a small value interpreter that cannot run arbitrary code. Hidden addresses, paths and method names are recovered and marked as decoded in the evidence.
5. Behaviour rules
Rule packs describe real attack behaviour, such as reading an access token and posting it to a webhook, or downloading bytes and defining a class from them. Benign patterns, such as showing your username in a menu, reduce the weight of a signal.
6. Score and verdict
Signals are combined into a risk score and a verdict. Clean mods stay low even when they use networking or reflection for legitimate reasons.
7. Explanation
An AI model can write a short summary of the findings from the static results and selected code samples. The decision is based on the analysis; the summary only explains it. See the privacy policy for what is sent.
Limits
- Static analysis cannot prove that a file is safe, only that nothing suspicious was found.
- Heavily customised or very new techniques may be missed.
- Server plugins are outside the main focus.
