Rubrical. Department for Education · Competition for Innovation winner
An AI marking and feedback platform for geography teachers, built to a government innovation contract and live in fourteen schools with the teacher holding the pen.
Department for Education Geography, KS3 to KS4 Fourteen schools live
Marking and feedback, built to a government contract.
OpenKit won a Department for Education Competition for Innovation contract to build a tool that uses generative AI to help teachers assess progress and give feedback on their students’ work. The Department is the customer and the product is ours, and delivery is reviewed against performance targets the Department set at the outset.
Rubrical starts with geography rather than trying to mark every subject, because a marking tool has to hold one subject’s mark scheme well enough for a department to defend the marks it produces at moderation. Development is complete, the closed pilot has run, and the platform is live in fourteen schools.
- 01
Funded by the Department for Education as a Competition for Innovation winner, delivered to its contract governance.
- 02
Every feature signed off by a teacher advisory board before it shipped, with the decisions documented.
- 03
Live in fourteen schools behind a trust-wide moderation framework, with marking time and teacher satisfaction still being measured as the rollout widens.
What teachers asked for first.
When we put a single expression-of-interest call through the geography teaching community, twenty teachers signed up inside a fortnight and eighteen of them named marking and feedback as the thing they most needed help with. Another eighty-two signed up to the two classroom-AI and Rubrical workshops we ran with our teaching partner. The brief set the constraints the platform was built to:
- 01
A teacher signs every grade. The platform drafts against the rubric and stops there.
- 02
The mark scheme the department already uses is the standard the model is held to.
- 03
Anything touching assessed work has to leave a moderation trail a school can produce on request.
Inside Rubrical.
- 01
A marking workspace the teacher signs off
The rubric and mark scheme go in, work comes in by upload or straight from Google Classroom or Microsoft Teams, and the platform drafts against each rubric criterion. Nothing reaches a student until a teacher has read it, edited it and approved it.
- 02
The PEEC+ analysis framework
A hierarchical model for handling the argument structures in Key Stage 3 to GCSE long-form answers, developed and validated as a deliverable in its own right and integrated into the marking pipeline. It is what lets the system reason about a long-form answer as an argument rather than as a bag of keywords.
- 03
A lesson plan generator
Built alongside the marking work because teachers kept asking for the other half of the job, and signed off by the advisory board with the marking interface.
- 04
Class-level cohort analytics
Topic-level strengths, gaps and common misconceptions across a class, so a teacher can aim the next lesson at what the cohort actually got wrong.
- 05
Refinements for SEND and EAL
The teacher advisory board asked for these, and they were built and in the product before the closed pilot started.
- 06
A teacher advisory board with a vote
Teachers on the board score what gets built next, and each design decision is written down with the reasoning behind it. Every feature in the platform was signed off through them before it shipped.
Nothing reaches a student until a teacher signs it.
Approve it, revise it, or send it back
The student’s answer sits on the left and the rubric level, the mark range and the descriptor sit on the right, with a written summary underneath explaining which parts of the answer earned what and where the evidence was missing. The teacher adds their own comment and either approves the assessment or sends it to be revised. The draft is a starting point a teacher edits, and the review screen is built to make editing quicker than starting over.
What teachers reported.
These figures come from Rubrical’s own tests with teachers rather than from an independent study, and the marking-time figure is the one the Department for Education went on to publish.
less marking time, reported by teachers in the developer’s tests and recorded by the Department for Education in its schools white paper.
teacher satisfaction with the marking and feedback the platform drafts for them.
more detailed feedback than teachers had time to write by hand, against the same set of work.
of marks matched the senior teacher’s grade on held-out GCSE geography responses, against 67% for generic AI tools scored on the same rubric.
The Department for Education records the marking figure on page 98 of its schools white paper, Every child achieving and thriving (February 2026): “Teachers report a 52% reduction in marking time in the developer’s tests.” The same entry records the Department’s initial £1m award in 2024 and a further £1m from Innovate UK.
Where Rubrical is now.
Rubrical is in production, and the model is refreshed against new marking data each quarter as the rollout widens.
- 01
Technical development is complete, the closed pilot ran in three schools at two hundred papers a week, and the platform is live in fourteen with a trust-wide moderation framework signed off and service levels in place.
- 02
Marking time, teacher satisfaction and adoption keep being measured as the rollout widens, and the impact report we are writing jointly with the Department for Education will carry that evidence.
- 03
The commercial model moved from individual-teacher subscriptions to department-level licensing, because the demand that came back was for whole departments and for Key Stage 3 alongside GCSE.
The build
- Fine-tuned subject-specific model
- Private RAG
- DfE content store training
- Azure UK tenant
- OAuth via trust IdP
- Google Classroom + Teams integration
- KILN evaluation framework
OpenKit certifications
- ISO 27001
- ISO 9001, UKAS-accredited
- Cyber Essentials
Controls on this project
- Operates to UK GDPR
- UK data residency
- DfE DPIA approved
More of the work.
Find your first workflow.
We start with a conversation, audit where AI actually pays back, and build the first automation into how your team already works. We reply within one working day.
Start the conversation AI Audit