Giving forecasting models a map of the data's hidden shape

Most time-series forecasters treat each series in isolation, missing structural relationships across the full population. TopoPrimer computes persistent homology and spectral sheaf coordinates once per domain, then injects this topological context as input tokens or a lightweight adapter for pre-trained models like Chronos and TimesFM. Tested across four public benchmarks, the gains hold equally in zero-shot and fine-tuned settings — suggesting topology captures something orthogonal to per-series training. The advantage is sharpest under stress: during peak seasonal demand, standard models degrade by up to 50% while TopoPrimer stays within 10% of baseline accuracy.