Claim and source
LOW RISK—HIGH RISK
Live check
ENGLISHAnalyze a news claim
Model output is a risk signal, not a fact check. Use source evidence before sharing.
Prediction history
Search and filter past checks. Entries never leave this device unless you export them.
Training
Browser baseline
Upload a CSV with text,label; labels must be real/fake.
BERT + CNN comparison
Train on Mac with scripts/train_bert_cnn.py, then import models/metrics.json.
No BERT-CNN metrics imported.
- Use balanced labels
- Check macro F1, not accuracy alone
- Review low-confidence mistakes
Metrics dashboard
Prediction distribution
Feedback quality
Model comparison
Settings
Language mode
Language is detected automatically; choose a display preference.
Privacy
Training data, model, history, feedback, and theme remain in this browser. Groq receives claim text only when you click its explanation button.