One heart lesson, end to end — asset, lesson, attempt, evidence
This demonstrator runs the whole SpatialLearn chain on a single Key Stage 3 objective. The 3D heart is generated by a component-constrained pipeline, validated structurally, converted into an assessed lesson, and every learner interaction is traceable back to the exact lesson version that produced it.
- 2D teaching image
- curriculum & component understanding
- GPU 2D-to-3D generation — the research core
- 3D candidates
- structural & educational validation
- interactive lesson
- learner attempts
- traceable evidence
What is proved here
- 8 reviewed components survive generation as separately named, selectable meshes.
- Structural validation is a gate, not a report: 6 checks run, current outcome passed.
- Assessment is deterministic — 3 tasks scored against a versioned answer key, with no generative model in the scoring path.
- Every learning signal traces to task, attempt and event, and is suppressed when the evidence is too thin.
Generation record
| Pipeline | heart-procedural-component-aware-0.2 |
|---|---|
| Constraint set | ks3.heart.structure.v1 |
| Compute | cpu-only-procedural-baseline |
| Candidate | rank 1, promoted, coverage 100% |
| Asset checksum | sha256:1913d54538b1c014667ba0f… |
Re-run notebooks/CoreWord_SpatialLearn_Heart_GPU.ipynb on a Colab GPU to replace these values with measured ones.
Claim boundary
No learned 2D-to-3D model has been trained or evaluated yet. The generator is a procedural, component-constrained baseline whose purpose is to make every downstream contract real and testable before a checkpoint exists.
The lesson also carries no human science review: review.reviewer says so in the published configuration, and nothing in this demonstrator should be shown to learners until a named subject reviewer has signed the geometry, answer keys and accessibility text.