Linda-Pro: local English AI-text detector (stylometry + 2 DeBERTa) - looking for hard test texts

An open-source English AI-text detector called Linda-Pro, built by developer Lindarixon as an ensemble of stylometry plus two fine-tuned DeBERTa models scored in roughly 300-word windows, flags plain AI text at 99.7% on the Chicago Booth benchmark at a 1% false-positive rate, dropping to 85% after a StealthGPT humanizer and reaching 92.6% on 889 texts from generators the models never saw. The developer published weak spots including 8.8% false positives on TOEFL exam essays (1.1% in precise mode), plus humanizers and short texts, and released gated weights free for research on Hugging Face with code and an evaluate.py script on GitHub. Lindarixon is soliciting adversarial and non-native-writer texts to break the detector and offering a hosted demo with file upload via DM.

Ensemble of stylometry + two fine-tuned DeBERTa models, scored in ~300-word windows, with a per-sentence heat map for mixed authorship. Results on the Chicago Booth benchmark 1% false positives : plain AI 99.7%, after a StealthGPT humanizer 85%. On generators the models never saw 889 texts : 92.6% flagged. Weak spots published too: TOEFL exam essays 8.8% false positives 1.1% in precise mode , humanizers, short texts. Weights gated, free for research : Lindarixon/Linda-Pro · Hugging Face https://huggingface.co/Lindarixon/Linda-Pro Code and evaluate.py for your own texts: GitHub - Lendarixon/Linda · GitHub https://github.com/Lendarixon/Linda I’m looking for adversarial and non-native-writer texts to break it. DM me if you want a hosted demo with file upload.