It was a simple MVP for personal OSS donations, and I have many considerations on how to evolve it and especially to prevent it from becoming a victim of Goodhart's Law at scale. Some of them:
1) Value and Risk scores shall include more metrics: dependencies, known funding, time since the last dev activity, active contributors, etc. A wider set of connected but relatively independent metrics is harder to fake. Also, it will help to exclude edge cases — for instance, I auto-donated to Pydantic (it's a great OSS), but such support is unlikely needed as they have raised $12.5M Series A from Sequoia this year.
2) Algorithmic does not mean automatic. While I see a strict, measurable, and ideally fully transparent methodology crucial for donating, it does not mean that all inputs shall be automatically generated. For instance, in the stock ETF world, one can generally rely on such metrics as "annual financials" for trading because they are annually audited (although it does not prevent fraud in 100% of cases). In the OSS world, data from trusted ecosystem actors can also be part of the equation.
3) Many guardrails are possible: limited budget per project, manual checks of top repos with the most anomalous changes in metrics. Also, if we target the sustainable maintenance of OSS the world relies on (I do!), then new projects (1-2 years) will unlikely get high scores - that adds another protection layer.
Given the interest in this topic, I am going to continue developing this algorithm further and expand it to other ecosystems (e.g. JS/TS and Rust). Your feedback here is very valuable to me, and those who would like to help make the algo better or donate through it are invited to the newly created gist:
https://gist.github.com/vinogradovkonst/27921217d25390f1bf5e...