What needed solving
A compelling 3D turntable rarely explains which parts were observed, inferred or produced from a synthetic camera assumption. That makes technical review and responsible reuse difficult.
How we're building it
We built a public workbench that accepts an image or small image set, predicts depth, projects geometry, fuses views, exports portable 3D artefacts and returns a manifest covering model, runtime, hashes, camera-pose source, confidence and fallback behaviour.
What we learned
Reproducibility is part of the product surface. Evidence cards and machine-readable manifests make a spatial result easier to inspect, reproduce and challenge without taking away from the visual experience.
What we are testing next
Which evidence fields best predict whether an operator can safely reuse a reconstructed asset in inspection, documentation or digital-twin workflows?
Where it is
- 2026Public GPU Space returned a complete 14-output evidence bundle
- 2026Hosted verification recorded a passing 78-test local regression gate
- NextEvaluate controlled real-capture sequences with named datasets and reproducible manifests