By this, we mean that every week, this repo should run the calibration routine on the latest version of the country packages, update the weights as a pandas CSV for each country in a data release on this repo, and run pytest tests that ensure:
- No negative weights
- Effective sample size (number of nonzero records) is larger than 10k
- Loss is reduced