Methods

Learning. The anatomy is a template and partly schematic, the simulations are simplified models, and nothing here describes a particular person. The site holds no patient data, collects no health data, and gives no diagnosis, prognosis, treatment advice or doses.

It is a free, non-commercial showcase: nothing is sold, there is no paid tier, no sign-up and no tracking. Content is licensed CC BY-NC-SA 4.0 (required by the OpenStax text it quotes); the source code is MIT. Every dataset, font and library it uses is free.

Bones. The vertebral column, sacrum, pelvis and body surfaces come from the HuBMAP Human Reference Atlas, whose united male model is derived from the Visible Human male. One subject: the Visible Human male, imaged as a supine cadaver. Not a population template.

Cord. The spinal cord, grey matter and white-matter tracts are the PAM50 template and atlas De Leener et al. 2018; Lévy et al. 2015. They were placed in the column by vertebral coordinate (PAM50 vertebral coordinate (body centres C1-L1) mapped to the subject's canal centreline). Where the template and the subject disagreed, the cord was moved off bone (at most 5.4 mm, mean 0.4 mm); no cord vertex lies inside bone.

Segments. Segment positions come from the template's spinal levels and published estimates Frostell et al. 2016; Cadotte et al. 2014. The largest disagreement between our placement and the reference atlas's own segment position is 13.0 mm, at S2.

Schematic layers. Roots, ganglia, spinal nerves, the dural sac, filum terminale, ligaments and vessels are drawn from rules, not traced from imaging, and are labelled schematic wherever they appear. Roots leave by the rule in the textbook Betts et al. 2022, §13.4. The dural sac's lower end is placed at S2, the most common level in an MRI study of 200 adults AL Jawad et al. 2022.

Frame. Visible Human male body frame (HRA united-male v1.9), RAS millimetres. three.js (x, y, z) = (R, S, -A), 1 unit = 1 mm. It is not MNI152.

Measurements. Measured on one supine cadaver (Visible Human male). Standing values differ. Not normative. Alignment angles shown in the lab are measured automatically on this one model.

The full-body skeleton, muscles, arteries, veins, organs, brain and skin come from BodyParts3D 4.0. Each part keeps its anatomical concept name; parts were sorted into layers by ordered rules over those names, and each layer was simplified per part with an error limit, keeping every part and moving no bounding box by more than 2 mm.

Each connection between the spine and the body names its spine end (vertebrae or cord segments), its targets, and the exact sentence from its source. The lines drawn between them are schematic: they show which ends are connected, not the course of a nerve or vessel.

• Every anatomical statement is a quotation, or a paraphrase placed next to the quotation it rests on, with its source. Sentences were copied from the source on the date shown.
• In articles, every paragraph and list item must cite a source, and every number with a unit must name its source; the build fails otherwise.
• Formulas are data: each has variables with units, a validity range, sources and a worked example the build recomputes.
• Facts measured on the 3D model say so, and say that they come from one person.
• Analogies and simplified models are labelled, with where the analogy holds and where it breaks.
• Anything we could not source is left out or labelled schematic, and listed here or in the source notes.

Research claims on the Frontier page and in articles carry a tier and an as-of date. The tier is shown as text and a shape, never colour alone.

• T1 peer-reviewed human study
• T2 peer-reviewed animal or in-vitro study
• T3 preprint
• T4 company announcement or press report
• T5 expert opinion or textbook convention

Every lab runs a model in a worker and labels each parameter as measured, a convention from a named paper, or a model choice. The neuron lab uses the Hodgkin–Huxley equations with their published constants Hodgkin and Huxley 1952; conduction times use Hursh's ratio Hursh 1939. The validation page runs checks against published values and numerical properties at build time and shows the results.

Raw inputs are pinned by checksum, and the full build is one command; see Data for the built files.