CyberScale Operational Severity v1

Incident operational severity classifier (O1-O4). Assesses consequence and coordination needs from a crisis management perspective.

Model Description

  • Architecture: ModernBERT-base with 4-class classification head
  • Training: 8,000 parametric incident scenarios (50 templates ร— field combinations)
  • Confidence: Monte Carlo dropout (20 passes) maps variance to high/medium/low
  • Labels: O1 (local) โ†’ O4 (EU-wide crisis)

Intended Use

Classify the operational severity of cyber incidents based on entity relevance, cross-border impact, member states affected, and coordination needs. Part of the CyberScale dual-scale incident classification system (T-level + O-level โ†’ Blueprint matrix).

Input format: <description> [SEP] sectors: <list> relevance: <level> ms_affected: <N> cross_border: <level> coordination: <level> capacity_exceeded: <bool>

Metrics

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

Citation

Part of the CyberScale project โ€” multi-phase cyber severity assessment MCP server.

Downloads last month
6
Safetensors
Model size
0.1B params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Evaluation results