# Static shoulder and spatial interaction candidate v0.7

14 September 2026. Anatomy App Workstream. Local candidate; not a replacement for the retained learning app or an anatomically approved shoulder model.

## Scope and changes

The user approved the static shoulder teaching milestone, middle-mouse dragging and separation that preserves relationships within muscle groups. The account reserve for this workstream is 50% remaining, superseding historical 5%/10% thresholds for this task.

This candidate derives from full-body v0.6 and loads unchanged v0.5 geometry. Older viewers, source masters, rigs, identities, material files and coordinate conventions are preserved. No distinct Exercise Library consumer is confirmed or changed.

- Hold the mouse wheel and drag to pan in X/Y. Middle clicking does not select anatomy or start browser autoscroll. Shift + drag and Shift + arrows also pan. Reset/focus clear pan. Left drag rotates; wheel scrolling zooms.
- Separation no longer assigns array-ordered grid slots. Above 10%, original 3D centre differences are scaled by a common positive factor. Relative size and orientation remain consistent, tissue parts receive identical transforms, and 1–10% remains a shared pull. Overlap from a particular viewpoint can remain; rotation reveals depth without arbitrary reordering.
- Supraspinatus, infraspinatus, subscapularis and teres minor have separate attachment descriptions, side-specific receiving-bone references and tissue-review status. Origin, Insertion and Both highlight scapula/humerus in blue/amber. During separation references stay with the context bones. These are not footprint overlays.

## Fresh source audit

`shoulder-source-audit.json` was generated by `tools/blender/audit_static_shoulder_v07.py` against the v0.5 Blender master. It checks actual triangle assignments, material seams, edge topology and browser-export correspondence. The master hash remained unchanged.

| Muscle, each side | Tendon triangles | Material seam edges | Interpretation |
| --- | ---: | ---: | --- |
| Infraspinatus | 521 | 39 | Source tendon/seam exists; anatomical boundary review pending. |
| Supraspinatus | 0 | 0 | A material slot does not supply an actual tendon region. |
| Subscapularis | 0 | 0 | A material slot does not supply an actual tendon region. |
| Teres minor | 0 | 0 | A material slot does not supply an actual tendon region. |

All eight meshes have zero boundary edges and zero edges shared by more than two faces under this audit. These limited topology checks do not establish anatomical shape or contact correctness. No material reassignment, new tendon mesh or footprint was invented.

## Unfinished shoulder acceptance requirements

Follow-on investigation found 16 source-native attachment patches in the original Blender file. Their receiving-bone correspondence and visual locations have been inspected in the separate `Research/2026-09-14/shoulder-patch-review.html`. See `SHOULDER_PATCH_REVIEW.md` there. This supplies a concrete mapping candidate; it does not yet establish the borders or repair missing tendon assignments. The teaching app does not load the research patches.

The approved shoulder milestone is **not complete**. Descriptions and a reference UI do not establish model-specific footprints or tendon boundaries.

For each of `ZA_SUPRASPINATUS_MUSCLE_{L,R}`, `ZA_INFRASPINATUS_MUSCLE_{L,R}`, `ZA_SUBSCAPULARIS_MUSCLE_{L,R}` and `ZA_TERES_MINOR_MUSCLE_{L,R}`, the remaining anatomical work must establish:

1. The muscle/tendon transition on the pinned mesh as reviewed face assignments or a supported replacement surface; review the retained 39-edge infraspinatus seam.
2. The origin area on `ZA_SCAPULA_{L,R}` and insertion area on `ZA_HUMERUS_{L,R}`, with reviewed faces/anchors and named landmarks.
3. Supporting evidence, reviewer/status, side and geometry revision/hash. Broad attachments must remain areas, not arbitrary dots.
4. Export and visual checks of both sides, rotation, labels, tissue continuity and separation after mapping. Movement remains outside this static milestone.

Current source assignments and references do not establish those exact face maps. `anchors` and `footprints` remain empty. Rig contact seeds, unverified `.or`/`.er` markers, bounds and nearest-bone distances are not substitutes. Published evidence also cautions against a simplistic supraspinatus/infraspinatus facet division; see `Research/2026-09-14/STATIC_SHOULDER_EVIDENCE.md`.

## Validation

- `qa-spatial-v07.json`: 15,040 cases and 368,800 pair comparisons across all existing multi-member course groups, both sides/bilateral, desktop/narrow, yaw/pitch, normal/isolated and separation stages. Reverses array order and checks relative 3D directions, first-stage spacing and common scale.
- `qa-browser-v07.json`: actual Edge WebGL panning, GPU picking, selection, isolation, hiding/restoration, staged separation, rotation, wheel zoom and narrow layout.
- `qa-shoulder-teaching-v07.json`: four muscles on right/left/both, all reference modes, correct receiving IDs, pending footprint status, tissue counts, separated references, labels and narrow layout.
- `qa-v07-*.png`: browser screenshots for visual inspection. Software tests are not anatomical approval or deployment.

One interaction-test attempt failed because its screen rectangle was captured before page scrolling; the test now reacquires it. A concurrent shoulder-test attempt timed out at initial loading; the separate final run passed. Preserve this limitation when interpreting performance; the checks do not establish a multi-browser GPU capacity target.

## Reproduction

Run `tools/blender/audit_static_shoulder_v07.py` through the FIT Anatomy Blender wrapper, then `tools/build_explorer_v07.py`. Canonical changes are in that generator and `tools/viewer-v07/`; generated-file edits would be overwritten. The source audit is required to generate teaching records.

Serve the project root on loopback with sibling v0.5 present. Current preview: `http://127.0.0.1:8773/Anatomy%20App%20Concept/Blender/musculoskeletal-v07/index.html`. No publication is recorded. Model notices remain in `ATTRIBUTION.md`; app code and FIT branding do not change model reuse conditions.
