TelecomTS

Summary

TelecomTS is a multimodal observability dataset from a controlled 5G telecommunications testbed. It pairs scale-preserving KPI time series with natural-language descriptions, anomaly metadata, labels, and Q&A fields for forecasting, anomaly detection, root-cause analysis, and time-series/text reasoning.

Official Artifacts

Dataset Shape

TelecomTS exposes 32k chunked samples on Hugging Face. The paper reports 1,020,000 normal observations, 120,000 anomalous observations, 18 KPI channels, 10 Hz sampling, 30,000 jamming observations, and 10 synthetic anomaly types.

Role In The Wiki

TelecomTS should be considered for experiments that need observability data with preserved scale, telecom-specific semantics, anomaly labels, root-cause labels, and language fields. It is a better fit for anomaly/root-cause/multimodal reasoning experiments than for pure passive forecasting leaderboards.

It is a near-miss for action-conditioned observability world models. Jamming, congestion, mobility, and synthetic anomaly events condition the data, but the dataset does not provide a logged operator action channel for counterfactual planning.