A public layer for Harmony XR+AI, spatial AI, and multimodal context inference while the deeper drafts stay private until the argument and evidence are stronger.
This lane keeps the public argument legible without exposing unfinished paper drafts too early.
How can XR systems perceive, remember, reason, and adapt across sessions instead of resetting each time?
A sharper Harmony architecture note with figures, terms, and claims that can survive a research review.
The current pass is about tightening the systems argument so perception, context, memory, adaptation, and explainability read as one architecture.
Gaze, head pose, hands, posture, boundaries, and task state matter only when the system can explain what changed and why it adapted.
Signals should map to task state, confidence, and interface decisions, not merely populate a dashboard.
Adaptive XR needs readable reasoning so users can audit changes instead of feeling managed by the system.
The next version should include a figure, demo clip, repo link, or concrete research question.