eromang/cyberscale-technical-v1

CyberScale Technical Severity v1

Incident technical severity classifier (T1-T4). Assesses observable technical impact from a CSIRT perspective based on structured incident fields.

Model Description

• Architecture: ModernBERT-base with 4-class classification head
• Training: 8,000 parametric incident scenarios (50 templates × field combinations)
• Confidence: Monte Carlo dropout (20 passes) maps variance to high/medium/low
• Labels: T1 (minor) → T4 (catastrophic)

Intended Use

Classify the technical severity of cyber incidents based on service disruption, affected entities, cascading effects, and data compromise. Part of the CyberScale dual-scale incident classification system (T-level + O-level → Blueprint matrix).

Input format: <description> [SEP] disruption: <level> entities: <N> sectors: <N> cascading: <level> data_compromise: <level>

Metrics

Metric	Value	Target
Accuracy	1.0	> 75%
Macro F1	1.0	> 75%

Citation

Part of the CyberScale project — multi-phase cyber severity assessment MCP server.