A verification pipeline for evaluating models and forecasts.

• 📥 Fetching data
• 🧮 Computing scores
• 📝 Writing results

• ✅ Full control over the verification pipeline via configuration
• ✅ Native integration with Delft-FEWS
• ✅ Builds on Scores for computation of scores. This package has extensive functionality, and it's documentation is world-class.
• ✅ Extensible with your own (private) datasources, scores and datasinks
• ✅ Optimized internal datamodel for efficient computation

Install from PyPI:

pip install veriflow

Or using uv:

uv pip install veriflow

See CONTRIBUTING.md for development setup.

This project is aimed at anyone who's interested in assessing model and forecast quality in an easy and reproducible way, like:

• operational forecasters
• model developers
• researchers and data-scientists

Verification pipelines are complex: metrics require specialized computation, data volumes may exceed memory, and sources/destinations can vary widely. This package simplifies verification by handling the entire pipeline via a single configuration file. It's reliable (tested and versioned), transparent (fully documented), reproducible, and flexible—extensible with custom datasources, scores, and datasinks. Any pipeline created is immediately transferable to other users and systems.

So wether you're working on model development or operational forecasting: this tool can help you build robust and reproducible verification pipelines.

• Builds on Xarray for handling multidimensional data.
• Supports Zarr for cloud-friendly data storage
• Supports Dask for parallel and lazy computation