Motion Accountability Platform — Architecture

### Vision Always-on observation layer on top of MotionMix's existing pose/avatar pipeline. Not a fitness app feature, but a platform primitive that uses continuous motion capture to detect exercise reps, classify sleep/wake state, and verify public commitments. ### Philosophy "Not direct features per se, but building on known observations that we want to highlight or improve or disincentivize." The system watches, classifies, and surfaces, rather than prescribing. The claim_bridge.rs already detects: - **Echo** (periodicity > 0.8): rhythmic repetitive motion = exercise - **Oscillation** (6+ curvature sign changes): bouncing/reps - **Dwell** (speed < 0.02 for 10 frames): pause between sets - **Transition** (jerk spike > 2sigma): rep boundary Events that trigger clip capture (via existing SANTrajectoryLogger): 1. Rep completion → 3-second clip centered on rep 2. Exercise bout complete → summary clip (first + last rep) 3. Wake-up event (Sleeping → AwakeActive transition) 4. Commitment verification moment | System | How It Connects | |--------|----------------| | Avatar Pipeline | Provides 27-bone positions at 30Hz via `avatar_get_bone_positions` | | SAN Claims | Bout segmentation uses Echo + Oscillation + Dwell signals | | PoseMetrics | Body energy, bounce, hip Y already computed | | Auto-Capture | Exercise bout/rep/wake events trigger clip recording | | Notifications | iOS local notifications for interval reminders | | Supabase | Commitments, sessions, sleep events persisted | | Feed | Commitment verification results posted to social feed |