For research
One coordinate system. Published methods.
Every figure is a deterministic transform of what a human painted, and the formula that produced it travels with the data.
Where the evidence stands
What has been characterised, and what has not.
The instrument-internal question — does the paint-to-metric transform add noise of its own — has been worked on in earnest. The human and cross-device questions have not been touched, and no amount of engineering answers them. Metrics are presented as derived symptom-burden measures, not as validated clinical endpoints.
- Characterised: a Monte-Carlo harness runs synthetic scenarios through the unmodified shipping metrics engine and reports ICC(2,1), SEM, MDC95, within-subject CV, and standardised response mean per metric.
- Characterised: a UV-seam artefact that split roughly 36% of single brush stamps at 1024² was measured on real GPU output and taken to zero by a world-space cluster merge.
- Not measured: human test–retest reliability. Whether a person repainting the same symptom produces the same figures is unknown.
- Not measured: cross-device and cross-body comparability. Whether a texel means the same anatomy on two different people is asserted by the shared atlas, not demonstrated.
- Not measured: validity against established clinical anchors.
- Not shipped, on purpose: magnitude bands and change thresholds, which need minimal-detectable-change studies that have not been done.
What you get
Records built to be re-analysed.
Each capture is a PAINprint Snapshot; an ordered sequence is a PAINprint Timeline. Both export losslessly, and both carry the compute contract that produced them — scoring version, body-model version, every formula, the area alpha threshold, the body-surface-area denominator, and the overlap policy.
- Canonical JSON: versioned, lossless, with the methodology block attached.
- Research CSV: finding-level rows for analysis without parsing anything.
- Deterministic throughout — no model, no inference, no learned weights, no hidden state.
- Any figure can be rederived by hand from the published contract.
What we do not have
There is no dataset to request.
Charts are stored in the browser that made them and never reach a server, so there is no pooled corpus here — not a small one, not a private one. There is also no research consent flow yet, and consent cannot be applied retroactively to charts already made. Any study therefore begins by collecting its own data under its own protocol, with its own determination in place before the first record.
- No server-side record store exists.
- No chart made to date carries research consent, so none can be released.
- A study collects prospectively, under its own IRB or non-human-subjects determination.
- That determination has to precede collection — it cannot be obtained afterwards.
The work outstanding
The order it has to happen in.
The gates below are sequenced because each one makes the next interpretable. Running them out of order produces numbers that look like findings and are not.
- Human test–retest on real repainting, to establish a noise floor.
- Cross-device and cross-body correspondence, before any figure from two people is compared.
- Minimal detectable change, which is what makes a delta readable at all.
- Convergent validity against established instruments.
Bring a question, not a purchase order.
The most useful conversations so far have started with a study design and a doubt. Write to research@3dpain.app if that is easier than a form. Cite mannequin variant and version when reporting absolute cm² figures.