Great question. I didn’t put any restrictions on the amount of time from publication to analysis. This may invalidate the results if there are significantly more newer stories that haven't had time to accumulate claps. (Or many old stories…)

I may come back to this in the future and only look at articles that have been published for the same amount of time. This should give an unbiased performance metric that doesn’t depend on time.

Many ideas coming to me now, thank you!