| name | inferred-motion-views |
| description | Build or refine the inferred motion layer, including synthetic front/rear/lateral renderings, model-confidence UX, inferred-motion DTOs, and safety language. Use this when working on 2D/3D motion rendering, replay, or advanced Wi-Fi pose features. |
Inferred Motion Views
Use this skill when the task involves:
- 2D keypoint rendering
- 3D skeleton rendering
- synthetic front / rear / lateral view UI
- inferred motion DTOs
- replay with inferred motion
- uncertainty overlays
- model-confidence presentation
- experimental pose features from Wi-Fi sensing
Objective
Support advanced motion visualization without implying the system is a camera.
These views are:
- synthetic
- model-based
- inferred from Wi-Fi-derived signals
- uncertainty-sensitive
- experimental unless validated
They are not:
- optical footage
- true camera views
- motion-capture truth
Mandatory user-facing language
Whenever synthetic motion is shown, include language equivalent to:
- "This is a synthetic model-based rendering inferred from Wi-Fi sensing."
- "It is not a true camera or optical motion capture view."
Also surface:
- confidence
- validation state
- signal quality
- model version where useful
View design rules
Allow view selection for:
- front
- rear
- left lateral
- right lateral
- orbit / free view when available
Make uncertainty visible with:
- opacity shifts
- confidence badges
- warning banners
- subdued styling for low-confidence frames
Data model rules
Store and stream inferred motion as structured inferred data:
- frame timestamp
- keypoint/joint schema version
- confidence
- model version
- experimental flag
- validationState
- source signal quality summary
Do not store it as camera media or label it as video.
Implementation guidance
When building inferred motion features:
- Keep the inference adapter separate from UI rendering.
- Keep the renderer separate from the model output schema.
- Provide typed DTOs for 2D and 3D outputs.
- Preserve model version and schema version.
- Add obvious empty, loading, and low-confidence states.
- Ensure replay timelines stay aligned with session timestamps.
- Keep the feature disable-able if the installation is not validated for it.
Scientific rules
Do not claim:
- exact joint kinematics
- exact foot strike geometry
- exact rear-view camera biomechanics
- optical equivalence
If the model is trained using cross-modality supervision, document that clearly.
Always update
When using this skill, review:
- docs/inferred_views.md
- docs/sensing_limitations.md
- apps/web inferred-motion screens
- apps/gateway inferred-motion adapters
- backend inferred-motion persistence and APIs
- tests for warning banners and DTO schemas