XENON MICRO-FRICTION v0.8
Developed by XenLogic
XENON MICRO-FRICTION is an experimental neural network for analyzing browser interaction event sequences, detecting possible interaction friction, and generating diagnostic reports.
Model specifications
| Property | Value |
|---|---|
| Version | 0.8 |
| Parameters | 3,000,175 |
| Architecture | Custom PyTorch GRU |
| Training examples | 5,000,000 synthetic records |
| Languages | English, Russian |
| Weights | Safetensors |
Classification
The model predicts three interaction states:
- NORMAL — expected or successful interaction
- CONFUSION — possible interaction difficulty
- BLOCKED — interaction appears blocked
Report generation
The model can generate English diagnostic descriptions and Russian-language reports based on event sequences.
Evaluation
Synthetic validation during training reported:
- Status classification accuracy: 100.00%
- Teacher-forced token accuracy: 96.40%
These are synthetic-data metrics and do not establish accuracy on real-world websites.
In an exploratory browser test involving 15 websites, the model predicted NORMAL for all 15. The browser instrumentation detected one JavaScript error on one tested website.
This indicates that further evaluation and calibration are needed before production deployment.
Limitations
- Experimental research model
- Trained primarily on synthetic event sequences
- May produce incorrect or contradictory reports
- May overestimate classification confidence
- Not a replacement for full browser functional testing
- Requires custom PyTorch architecture code for inference
Developer
XenLogic
Project: XENON MICRO-FRICTION
Version: 0.8
- Downloads last month
- 59