DualWeaver: Synergistic Feature Weaving Surrogates for Multivariate Forecasting with Univariate Time Series Foundation Models

DualWeaver addresses a known limitation in Time-Series Foundation Models (TSFMs): most are trained as univariate (channel-independent) models to maximize generalization, which causes them to miss cross-variable correlations in multivariate data. While the 'surrogate series' approach is a clever architectural tweak to inject cross-channel information, the project currently lacks any significant adoption (0 stars, though 6 forks suggest academic peer interest). It is highly vulnerable to frontier lab displacement; companies like Salesforce (MOIRAI), Amazon (Chronos), and Google (TimesFM) are already iterating on multivariate extensions. If any of these labs release a native multivariate update, the need for a 'weaving' surrogate layer like DualWeaver effectively vanishes. The project is currently a research artifact with low defensibility and a very short displacement horizon as the field converges on unified multivariate/univariate foundation models.