Hacker News Reader

They can automatically determine which are their own model outputs anyway by comparing the code to their internal logs

Are they practically doing this ?

When it comes to data quality it justifies a few checks.

Seems impractical