Frequently asked questions
Does the app send my texts anywhere?
No. Analysis runs locally and texts are never sent. The app downloads update lists and update files only.
What do I need to run it?
Windows 10/11, 64-bit. No GPU is needed. Current Windows includes the WebView2 runtime used by the app.
How many documents should I load, and how long should they be?
For document comparison, load 4–200 documents in one language, each at least 300 words; about 1,000 words is better. For authorship checking, use at least two known samples per author and at least two authors.
Is a high score proof of authorship?
No. Results are relative style distances within the supplied corpus. They do not prove authorship, identity, authenticity or leaks and do not replace expert examination. Candidates are not probabilities.
Which languages and file formats are supported?
English, Russian and Polish. TXT, DOCX, PDF, RTF, HTML and Markdown. TXT supports UTF-8, UTF-16 and selectable Windows-1251/1250/1252.
How do updates work, and can I turn them off?
The Updates button checks an Ed25519-signed list on Hugging Face. Model packs are SHA-256 checked and applied atomically; app updates are verified and installed from GitHub Releases. Turn off automatic startup checks in the Updates window.
Why does Windows SmartScreen warn me?
The installer is not code-signed yet. If SmartScreen warns, choose “More info”, then “Run anyway”; compare the published SHA-256 before running it.
Where are saved datasets, and how do I remove them?
Cases are stored on your PC in %LOCALAPPDATA%\ForensicAuthorship. During uninstall, the app asks whether to delete them.
What data was the model trained on?
The bundled resource has 2,000 character 3–4-gram features per language, trained only on documented public-domain and CC0 literary prose: 407 works.