Welcome to the Zoltar forecast archive, an open-source web application that facilitates the storage, retrieval, evaluation, and visualization of point and probabilistic forecasts. Zoltar was developed to assist with many kinds of real-time forecasting projects.
Designed to improve the robustness of forecasting research, Zoltar is a research data repository that stores forecasts made by external models in standard formats and provides tools for validation, visualization, and scoring. It builds off of a foundation of core ideas and data structures first introduced in 2019 by predx. Zoltar can host real-time or retrospective forecasting challenges, competitions, or research projects, with users specifying the forecast targets.
In June 2020, we released a preprint describing the concepts, data model, and intended scope of Zoltar.
Projects are the fundamental organization unit of Zoltar, and hold a collection of models and their forecasts. Project owners can customize the details of a project for a forecasting challenge, a collaborative research project, or a teaching workshop.
(Yeah, I don't know, either.)