
The model that writes your content cannot grade it
An early version of our AI-generated grammar exercises was confidently wrong 15% of the time while the answers were still correct. This is the pipeline that catches that, and the number that decides whether you can ship.




