CastWorld: Agentic Benchmark for Context-Aware Time Series Forecasting
November 13, 2026
A new benchmark featuring 25 real-world datasets across 8 domains to evaluate autonomous forecasting agents. It includes a modular harness with 49 tools and a state-adaptive loop for iterative evidence gathering and forecast revision.
HOW THIS AFFECTS YOU
●
builderYou can use this framework to build more robust agents that react to new data in time-series tasks.
●
researcherYou can now test forecasting agents on how they use external context and iterative revision rather than just one-shot accuracy.