brain backup 2026-09-14

projects/factory/family-voice.md

@@ -0,0 +1,45 @@
+---
+project_key: family-voice
+status: active
+factory_maturity: discovery
+updated: 2026-08-17
+---
+
+# Family Voice — Project Definition
+
+## Purpose
+
+Family Voice covers the privacy-first local voice-context and speaker-identification prototype, with future device integration only after offline evaluation and explicit consent decisions.
+
+## Execution map
+
+| Host / target | Repository / service | Paseo-project role | Status |
+|---|---|---|---|
+| Mac mini `.102` | `~/Projects/voice-speaker-id` | `Family Voice · Speaker ID` | Registered local workspace `wks_42151868896725e9` on `main`; offline CLI-only prototype |
+| ESP32 / iPhone devices | audio input sources | Project assets, not Paseo hosts | No always-listening deployment is approved |
+
+## Personas
+
+- Privacy planner: evaluates scope, consent, retention, and data boundaries.
+- ML experimenter: isolated code work without raw-audio export.
+- Evaluation verifier: runs reproducible offline scoring/evaluation only.
+- Privacy/security reviewer: checks data handling, authorization claims, and accidental cloud transmission.
+- Device integration planner: design-only until explicit device approval.
+
+## Safety boundaries
+
+- Cached clips are never treated as approved enrollment data by filename alone.
+- Do not upload raw audio, commit recordings/embeddings, create a listener, modify firmware, or enable always-listening capture without explicit approval.
+- Speaker recognition must never be the sole authority for high-impact actions.
+
+## Factory readiness next actions
+
+1. Confirm repository checks and ignored-data boundaries.
+2. Register the local repository workspace.
+3. Define an evaluation-only Paseo contract.
+4. Run no automated collection/deployment schedule until consent, retention, and calibration decisions are documented.
+
+## Related records
+
+- `../family_voice_context_and_speaker_id_2026.md`
+- `../../areas/operations/services_inventory.md`