ChronoGraph
Summary
ChronoGraph is a graph-native multivariate time-series benchmark from production microservices. It combines service node metrics, directed service dependency edges, temporal edge features, and expert incident labels.
Official Artifacts
- Preprint: https://arxiv.org/abs/2509.04449
- Official code and data: https://github.com/bit-ml/ChronoGraph
- Dataset metadata snapshot: chronograph-2025
Dataset Shape
ChronoGraph reports 708 services, 1529 directed edges, five node features, eight edge features, 8005 aligned time steps, and 17 incident segments.
Role In The Wiki
ChronoGraph is the clearest “whole graph plus multivariate time series” dataset in this comparison. It should anchor passive graph/time-series forecasting and anomaly-detection experiments. It should not be treated as action-conditioned because it has incident labels but no logged operator action channel.