Virtual patient core
Scenario banks, actor state, patient notes, review governance, and model routing are built as testable contracts before any runtime is treated as clinically ready.
Open clinical simulation infrastructure
OpenClinXR
Evidence-gated XR clinical training for browser and Quest workflows, with local-first voice, virtual-patient state, and production asset readiness tracked as explicit proof lanes.
Quest 3
USB and browser smoke gates
Headset evidence remains claim-scoped until foreground frame pacing is captured.
Voice
Local model and transport lanes
Moshi, Qwen-TTS, Bun, and Python proxy evidence stay separated from production claims.
Assets
Production evidence ladder
Contract fixtures are useful, but artifact-backed character and equipment proof is still gated.
What it is
A clinical XR stack that refuses vague readiness claims.
XR runtime evidence
WebXR and IWSDK sidecars capture browser, Quest, controller input, visual QA, and mixed reality evidence without blurring prototype behavior into production promises.
Local-first voice path
Local model caches, Qwen-TTS, Moshi package evidence, API websocket transport, and live-dialog benchmarks are verified independently and kept offline by default.
Evidence gates
Every capability carries its own proof boundary.
Automated asset generation pipeline
Local AI voice model strategy
IWSDK and IWER evidence runbook
Quest 3 USB WebXR smoke checklist
Full evidence documentation lives in docs/openclinxr/.
Run pnpm pages:sync-evidence-links and pnpm pages:validate to keep snapshots current.
Architecture
Decisions are recorded as MADRs.
Key technical and process decisions are captured as Markdown Architecture Decision Records (MADRs) in the repository. This provides a living, searchable history of why choices were made.
Examples include domain modeling, local-first voice strategy, Quest WebXR evidence gates, and asset pipeline design.
Development approach
Agent-augmented, evidence-first workflow.
OpenClinXR uses structured agent personas and skills (in /agents/) alongside MADR records
and automated evidence gates in /.agent-factory/. This keeps development honest, auditable, and aligned with clinical-grade rigor even in early stages.
Live snapshots
Current evidence posts are committed under one folder.
Production asset evidence ladder
Artifact-backed production asset proof manifest
Production asset readiness benchmark
GitHub Pages posture and validation notes
These snapshots are maintained via repository automation. See the pnpm pages:* scripts in package.json for sync and validation commands.
Current posture
Promising, deliberately gated, and still honest about blockers.
OpenClinXR is early-stage infrastructure, not a clinical product. The repo favors permissive runtime licensing, local/offline development evidence, and explicit blockers for physical Quest performance, production-grade generated assets, clinical validation, safety, and governance review.