Hacker News Reader

You can easily do a statistical signicance test to estimate if the effect is non-zero.

Which will still mean almost nothing, since past performance doesn't guarantee future results.

You basically just claimed that data analysis as a whole is invalid.

It's a valid point, this is post-hoc data analysis so you need to be careful of the Texas Sharpshooter Fallacy: https://en.wikipedia.org/wiki/Texas_sharpshooter_fallacy