Submit evaluation results for TimeRSI

TimeRSI results on the TIME benchmark: 98/98 tasks, each with config.json, metrics.npz, and predictions.npz containing quantile forecasts from 0.1 to 0.9.

TimeRSI combines a frozen TimesFM-3.0 backbone with code-agent-developed forecasting strategies and automatic selection. The selector may use feedback from earlier completed forecast windows, but never labels from the window being predicted. This is an agentic submission; no individual forecasts were manually selected or edited.

Local TIME aggregation of the submitted results gives normalized MASE 0.630011 and normalized CRPS 0.527713. We will release the agent at TimeRSI on GitHub.

Thank you for maintaining TIME.