The Same Anatomy, Four Ways to Understand It

Human Atlas becomes easier to judge when one named structure is followed across the boundaries already established: source meaning, preserved mesh identity, batched rendering, and changing scene state. This checkpoint uses that connected model to answer a practical question: why can assembled, layered, exploded, and isolated views show the same anatomy without becoming four different atlases? The answer is that the identity of the selected concept and its member pieces remains stable while the visual relationship among those pieces changes for a specific purpose.

One named concept, many preserved pieces

Start with a search result rather than a camera gesture. A named concept is a semantic entry point: it gives the reader a word and an atlas identifier that can be matched to an elements collection. The collection may contain one mesh or many. A selection of the concept therefore does not replace the underlying parts with an invented compound object. It selects the member IDs that the manifest already associates with that name.

That distinction makes a concept useful at two scales. At the broad scale, its name gives the reader something more meaningful to ask for than a raw mesh identifier. At the detailed scale, its elements keep the answer inspectable. The reader can know that a search result refers to a named structure and still see that the result contains multiple modeled pieces. The name organizes attention; it does not erase geometry.

This is why the detail surface can show both a structure name and a selected-piece count. The count is not an implementation curiosity. It tells the reader whether the chosen semantic unit maps to one visible element or a collection. An organ-scale concept can be represented by many source meshes, while a smaller or simpler concept may map to one. The interface preserves that difference instead of making every search result look equally indivisible.

The same rule applies when selection starts in the scene. A pointer can reach one source part, even when that part is displayed inside a merged system batch. The application can then present that part through the same details path used for a concept selection. The entry points differ—one begins with geometry, the other with a name—but both converge on an identity that can be described, selected, and isolated.

This convergence is only possible because the release preserved the link among concept records, part records, and geometry ranges. The source-to-browser transformation changed coordinates, triangle density, and storage format, but it retained the relationships needed to move from a name to member pieces and from a hit to a particular part. The runtime architecture then adds rendering state without changing those relationships.

The diagram describes a chain of relationships, not a claim that every semantic concept is one system or one chunk.