can you share the recipe for these values?

You can find it here: https://gitlab.com/dataspekt/hred2020 The 'fitted' properties (mean + p95 upper and lower bound) are calculated off of 2000-2019 data.