The classroom, rebuilt so a human keeps the grader's seat.
Solstra is a concept prototype of the layer a school actually runs on — units, live teaching, marking, a gradebook — reimagined so AI proposes at speed and a named human decides every judgement that reaches a learner. The responsible-AI guardrails live inside the platform, not bolted onto a side-tool.
Google Classroom owns the layer. This reimagines that layer — AI-native.
The tools that entered classrooms recently sit beside the system a school runs on, as a teacher's side-tool. Solstra rebuilds the system itself, AI-first, as a working concept: the model drafts, clusters and flags; the teacher stays the marker of record; and the trust controls are the point, not the fine print.
Everything a classroom needs — and the things Google Classroom can't do.
Live checkpoints
The class hits a mastery checkpoint; the model groups the wrong answers into proposed patterns. You review, approve a 60-second re-teach for just those pupils, and decide to progress the class.
AI proposes · you decideHuman-first marking
You mark against the rubric before the AI's suggestion is revealed — and you can't sign without your own mark. When the model reads a wrong answer as fine, the human is what catches it.
Marker of recordAssistants
Teacher-built, subject-specific helpers with guardrails baked in — hints not answers, on-topic only, and "that's one for your teacher" when out of scope. Under-13 is teacher-mediated with no pupil accounts.
Guardrailed · on ClaudeGrounded notebook
Answers only from the teacher's approved sources, cites which one it used, and when the answer isn't there it says so. Grounded, or silent — no open-web guessing near children.
Cited · source-onlyClass stream
Teachers post announcements and materials; pupils ask and respond. Nothing a pupil posts reaches the class until a teacher approves it — human-gated by design.
Human-gatedTrust Ledger
Governance as the architecture, not a buried PDF: human-marker-of-record and shadow-mode locked on, pseudonymisation and the under-13 rule enforced, a DPIA gate, and a session log of every step.
Locked on in the prototypeThe guardrails are the architecture.
Solstra is built so that improving model capability expands what AI assists with while the human-accountability core stays exactly where it is.
- Human as the marker of record — the model never awards a grade.
- Shadow-mode only — AI runs alongside human marking, never on a live grade.
- Pseudonymised inputs — names never reach the model.
- DPIA-gated — any live pupil-data use is blocked until the paperwork is done.
- Under-13 pattern — no pupil AI accounts; teacher-mediated throughout.
- Grounded, or silent — sources and citations, never open-web guessing.
Built by a computing teacher who runs this layer every day.
Solstra is an independent concept by Saqub Hussain — author of a complete, exam-mapped Reception-to-Year-11 computing curriculum published openly at plantbotcomputing.co.uk, and the builder of the working prototype. It runs a real Year 5 unit end to end. The delivery layer isn't a theory here — it's a daily practice, rebuilt the way it should be.
The whole concept is interactive — and it runs on gated, real Claude.
Open the classroom, run the Gate Rescue, catch the confident-wrong mark, ask the grounded notebook a question it has to refuse. Every claim on this page is one click away.