Here is a grade: 82%. Now do something with it.

You can't, really. You know you're somewhere above the middle. You don't know which eighteen percent went missing, whether it was one concept failing repeatedly or six unrelated slips, or what you should open first tonight. The number is a summary of a conversation you never got to have.

That's not an accident. Grades are compression, and compression is the entire point when the audience is a transcript, a registrar, or an admissions office. One number travels well. The trouble starts when we hand that same number back to the person who has to improve, as if it were feedback. It isn't. It's a receipt.

Where the information actually lives

The useful information was there and got thrown away. Consider two ways of reporting the same lost points:

• The receipt: "Question 4: 2/5."
• The diagnosis: "You set the integral up correctly and chose the right substitution, then didn't update the limits of integration when the variable changed. The same omission appears in question 7."

Same lost points. The second one is a study plan. It separates what worked from what didn't — which matters, because a student who reads "2/5" often concludes they don't understand integration at all, and re-studies the part they had already mastered. It names the operation that failed, so there's something to practice. And it links two incidents into one pattern, which is the difference between a careless mistake and a habit.

A score ends a conversation. A diagnosis starts one.

Rubrics, and why we grade against them

The way you get diagnoses instead of receipts is to decide, before looking at any student's work, what the work has to do. Not "how good is this" but a list: does it state the assumptions, is the method appropriate, is each step licensed by the one before it, does the final claim answer the question that was asked.

Grading against an explicit rubric changes the shape of the output. Instead of a holistic impression compressed into a number, you get a set of specific judgments — one per criterion, each attached to a place in the work. The score becomes a byproduct of those judgments rather than the thing being produced. You can still hand the number to a transcript. But the student gets the list.

It also does something quieter that we care about a lot: it makes the grader accountable. A holistic score is unfalsifiable — you can't argue with a vibe. A criterion-level judgment says exactly what it claims and points at the line it's claiming it about. If it's wrong, you can see that it's wrong. That's a feature, especially when the grader is a machine.

What we do with it

In Lune Synth™, every mark has to land on a step. The grader's job isn't to arrive at a percentage — it's to walk the work you actually wrote, find the first place the reasoning stops being valid, say what rule was violated, and keep going to see whether the same failure recurs. What comes back is a list of specific claims about your specific page.

Then those claims have somewhere to go. An error that shows up twice becomes the seed of a practice mission. A criterion you keep missing across different topics — say, checking domain restrictions, or justifying a step you treated as obvious — becomes a thing the app knows to watch. The grade stops being the end of the assignment and becomes the input to the next one.

None of this makes an 82% feel better. It just makes it mean something.

Want feedback that names the step instead of the score? Join the beta waitlist on the Lune Synth home page, or reach us at griffin@lunesynth.com.