Search papers, labs, and topics across Lattice.
ARFBench, a new benchmark for time series question answering (TSQA), is introduced to evaluate multimodal foundation models on software incident data. The benchmark comprises 750 questions across 142 time series derived from 63 real-world production incidents at Datadog, totaling 5.38M data points. Experiments reveal that frontier VLMs like GPT-5 achieve significantly better performance than existing baselines, and a hybrid TSFM + VLM model, post-trained on a small dataset, achieves comparable results, while a model-expert oracle demonstrates a new superhuman performance level.
Even GPT-5 only achieves 63% accuracy on time series anomaly questions from real software incidents, but a model-expert combination reaches 87%, highlighting the potential for hybrid intelligence in incident response.
Time series question-answering (TSQA), in which we ask natural language questions to infer and reason about properties of time series, is a promising yet underexplored capability of foundation models. In this work, we present ARFBench, a TSQA benchmark that evaluates the understanding of multimodal foundation models (FMs) on time series anomalies prevalent in software incident data. ARFBench consists of 750 questions across 142 time series and 5.38M data points from 63 production incidents sourced exclusively from internal telemetry at Datadog. We evaluate leading proprietary and open-source LLMs, VLMs, and time series FMs and observe that frontier VLMs perform markedly better than existing baselines; the leading model (GPT-5) achieves a 62.7% accuracy and 51.9% F1. We next demonstrate the promise of specialized multimodal approaches. We develop a novel TSFM + VLM hybrid prototype which we post-train on a small set of synthetic and real data that yields comparable overall F1 and accuracy with frontier models. Lastly, we find models and human domain experts exhibit complementary strengths. We define a model-expert oracle, a best-of-2 oracle selector over model and expert answers, yielding 82.8% F1 and 87.2% accuracy and establishing a new superhuman frontier for future TSQA models. The benchmark is available at https://huggingface.co/datasets/Datadog/ARFBench.
CMU ML