# v0.3.1 completion report ## Milestone **PersonaEcho Hospital Factory v0.3.1 — First-Person Onboarding & Hospital Integration Closure** The milestone closes the seven software P0 findings identified in v0.3. It produces a repeatable hospital-controlled workflow for one directly consenting adult and can be repeated for later people without changing the core software. ## Closed software P0 items | P0 | Closure | |---|---| | Speaker enrollment | versioned enrollment candidate, embedding hash, source/session binding, independent identity review and revocation | | Segment review | playable target-speaker clips, corrected transcript and explicit TRAIN/HELD_OUT/REJECT split | | Long-running processing | persistent asynchronous jobs, progress API, cancellation, restart recovery and bounded attempts | | Model data exchange | hash-bound shared-volume and S3/MinIO transfer with SigV4 probe and deletion | | Hospital authentication | OIDC RS256, gateway HMAC, replay protection, role enforcement and reviewer independence | | Deployment closure | Docker/systemd/Nginx examples; FFmpeg, OCR, PDF and malware-tool checks; real model probes | | Interface and handover | expanded OpenAPI, deterministic delivery manifest, software receipt and end-to-end hospital sandbox | ## Software evidence The repository verification checks JavaScript syntax, JSON/JSONL parsing, Git whitespace, production configuration, deployment dependencies, API route presence, secret patterns, tests and deterministic file hashes. The end-to-end sandbox covers: ```text project + consent → source upload and use approval → asynchronous speaker enrollment → independent identity review → asynchronous source analysis → human clip/transcript review → asynchronous candidate build → style/knowledge and dual voice review → release → chat → one-time speech → immediate revocation ``` The sandbox uses synthetic inputs and test model services. It demonstrates software state transitions and security controls; it is not evidence that a real person's voice or style has been faithfully reproduced. ## External activation gates that remain Software completion does not waive the following physical and institutional evidence: - exact license-cleared model weights and tokenizers; - real speech, LLM and TTS inference on the target hardware; - lawful source material and direct, revocable subject consent; - held-out identity, voice, style and knowledge acceptance; - hospital IAM, KMS/HSM, SIEM, network, backup, deletion, vulnerability and penetration-test acceptance; - accountable privacy, security, legal and release-owner sign-off. Until these gates pass, status remains `SOFTWARE_VERIFIED_REFERENCE_PRODUCT`, not hospital production acceptance. ## Deployment boundary The bundled store and queue support one active application instance for a controlled pilot. Multi-replica production requires transactional storage and distributed queue/token adapters followed by renewed acceptance testing.