Skip to the content.

RCA Ground Truth Labels

The supervision signal for GNN training — what the model learns to predict.

Source File

Label Schema (5 classes)

Label ID Name Count Nodes
0 root_cause 1 AGG-NYC-E-01
1 primary_symptom 3 CSG-NYC-E-01, gNB-NYC-E-047, eNB-NYC-E-047
2 secondary_symptom 3 CORE-RTR-NYC-01, UPF-NYC-01, AMF-NYC-01
3 collateral 3 gNB-NYC-E-048, CSG-NYC-E-02, AGG-NYC-E-02
4 unaffected 2 gNB-NYC-E-049, CSG-NYC-E-03

Training Target Formats

The labels are provided in three formats for different GNN tasks:

  1. Node classification (5-class): [1, 3, 4, 1, 1, 3, 4, 0, 3, 2, 2, 2]
  2. Binary root cause detection (2-class): [0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0]
  3. Root cause ranking (regression): scores from 0.0 to 0.95

Propagation Graph

A directed causal graph showing how the fault propagated from AGG-NYC-E-01 through 9 edges to 9 downstream elements. See rca-labels.json propagation_graph field.

Label Source

Derived from trouble ticket INC00847291, validated by senior transport engineer.