AI grading is only as good as the rubric behind it
In the largest meta-analysis of feedback, 38% of interventions made performance worse. What that means for AI grading, and for the rubric behind it.
Almost every AI education pitch cites Bloom's two sigma. The study ran three weeks and has never been replicated. What personalised learning can really do.


In the largest meta-analysis of feedback, 38% of interventions made performance worse. What that means for AI grading, and for the rubric behind it.

A fast first pass that returns in seconds, a bounded background fan-out to the model, and live grading progress streamed to the browser.

Every piece of student work becomes a targeted next step, for every learner, every week. What that looks like at the scale of a real class.

Forty essays can swallow a weekend. What changes when the first pass is drafted before you sit down, and why the marking still has to be yours.