CastFSR Framework Uses Agentic Reasoning for Time Series Forecasting

CastFSR introduces a Fast-Slow-Reflect workflow for context-aware forecasting. It uses lightweight models for initial data-driven priors, retrieves contextual evidence for deliberation, and employs a reflection step to validate forecasts against temporal and domain constraints.