Ribbon: scalable uncertainty quantification without repeated model refitting

arXiv·medium signal

This arXiv paper (2026-06-25) introduces Ribbon, a scalable approximation to Dirichlet-reweighted bootstrap uncertainty that replaces the expensive repeated refitting of bootstrap and fully-Bayesian methods with an influence-function linearization. It targets reliable predictive uncertainty for complex, high-dimensional, or misspecified models at a fraction of the compute. For practitioners, it's a practical path to calibrated uncertainty estimates on models too large to bootstrap.