Skip to content
coeducate

AI grading tool

AI grading tool that shows the reasoning behind the score

Marking speed was never the real cost. The cost is working out why each student got it wrong, which is the part that gets skipped when the stack is 140 deep.

Try the demo

01 Grading is two jobs

One of them is cheap and one of them is not

Deciding whether an answer is right takes seconds and can be automated by a multiple choice question. Working out what the student was thinking when they got it wrong takes minutes, cannot be automated by a rubric, and is the only part that changes what happens next.

Most tools sold as AI grading automate the cheap half faster. They return a score and a generic comment, and the teacher still has to open the paper to find out anything useful. That is why the time saving never quite materialises.

Coeducate does not return a grade at all. Grades belong to your grading policy and to the person who signs the report. What it returns is the expensive half: the diagnosis, the evidence and the mastery read.

Not returned

Score

A number. You already have a way of producing this and it is not where the hours go.

Not returned

Generic comment

Well structured, work on your explanations. True of everyone and actionable for nobody.

Returned

Named misconception

The specific move in the reasoning that fails, quoted from the work itself.

Returned

Per-skill mastery read

Three or four skills with a state, a percentage and an evidence line each.

Returned

Reteach activity

Five minutes, three steps, aimed at the diagnosed fault rather than the topic.

Returned

Report-ready sentence

One line describing what the student can and cannot yet do, ready to paste.

02 Checkable, not trusted

Every claim carries its evidence

AI grading will sometimes be wrong, in the same way a tired human grader at paper 96 is sometimes wrong. The difference that matters is whether you can tell.

A bare score is either trusted or ignored. A claim with the student's own sentence underneath it can be checked in about five seconds and overridden with one click, which is the only version a professional should accept.

Claim

Inferring author stance: Not yet, 18%

Evidence

"Attaches it was a good decision, which the passage does not say."

Source

Student answer, sentence 4. One click opens the full submission with that sentence highlighted.

If the teacher disagrees, the override is recorded against the skill and the series updates. The professional judgement wins, and the record shows that it did.

03 What it will not do

Three things we deliberately refuse to automate

Assign the grade

A grade is a decision with consequences for a student, made under a policy that belongs to the school. The tool supplies the input and stops there.

Send feedback straight to students

Unreviewed feedback is either bland enough to be useless or wrong in a way that costs more to unpick than writing it yourself. A teacher approves before anything reaches a student.

Detect AI use

Detectors produce false positives, disproportionately flag second-language writers, and an accusation you cannot substantiate costs more than the misconduct. We do not sell one.

The assignment-design alternative to detection is set out in writing a classroom policy that holds.

04 Where the hours go back

140 papers, five sections, one teacher

Check the arithmetic against your own schedule. If diagnosing one piece of work properly takes two minutes and you have 140 of them, the honest weekly cost of doing it well is nearly five hours, which is why almost nobody does it well.

This is not a claim about a measured result. It is arithmetic you can check with your own class sizes.

140

pieces of work per week, five sections of 28

2 min

to diagnose one properly rather than mark it

4.7 h

per week, which is why the diagnosis gets skipped

What comes back is not the grading time. It is the diagnosis that was never happening, which is the part that changes tomorrow's lesson. See how that feeds the next plan.

05 Pricing

Priced per program, not per seat

Annual billing is about 17% lower. There is no free tier, and the privacy posture is identical on every plan.

Classroom

$89/mo

One teacher, up to 60 students, CSV mastery export.

School

$349/mo

Up to 10 educators and 600 students, roles, SSO, scheduled reports.

District

$1,290/mo

Up to 75 educators and 5,000 students, standards mapping, SIS export, PO billing.

Mark one answer the expensive way

Paste a piece of student work into the demo and read what comes back. It is the diagnosis, not the score.

Student work is never used to train models. No card required to try the demo.

Try the demo

Mastery evidence after every session.