De-identified development
Development uploads require an explicit de-identification confirmation. Production use with PHI is not represented as ready.
Security and privacy
Trust starts with stating the boundary honestly.
OrthoCode AI is a development-stage system for de-identified charts. These are the safeguards implemented today—and the production work that remains.
No silent coding fallback
A JEV timeout, provider error or malformed response forces a RED review result without fabricated coding lines.
Licensed source isolation
Raw licensed codebook files remain outside the runtime API and frontend; runtime lookup uses normalized published releases.
Secrets stay server-side
The frontend receives no model-provider secrets or raw reference files. Provider permissions are explicit configuration.
Append-only review trail
Human corrections are added to the audit record without rewriting the original model output.
Autonomy guarded
Autonomous coding is disabled by default and remains gated by engine mode, successful execution, thresholds and rule results.
Before production
Controls still requiring formal review.
Production readiness is not implied
Production deployment still requires a compliance review, appropriate vendor agreements, approved private object storage, authentication and role-based access, key rotation, retention enforcement and audit monitoring.
Current posture in one sentence
Bounded de-identified inputs, no silent fallback, deterministic hard rules, mandatory human review, and an append-only correction trail.