TrajectoryOS Architecture

TrajectoryOS Architecture

System Overview

TrajectoryOS is a life-trajectory modeling system composed of cooperating services that infer your skills, alignment, constraints, and escape velocity through continuous interrogation and artifact analysis.

Core Principles

1. Latent State Model: Your life is represented as a time-series of latent vectors `z_t` from which all observable quantities derive
2. Bayesian Inference: Skills and competencies are probabilistic beliefs updated through evidence
3. Physics-Based Metaphor: Thrust, Alignment, Gravity, Mass, and Escape Index provide interpretable metrics
4. Future Embodied Integration: Designed to consume Echelon's movement dynamics when available

Architecture Layers

### Layer 1: User Interface
- web-dashboard: Next.js dashboard for visualizing trajectory, skills, and escape index
- api-gateway: HTTP REST + WebSocket server for sessions and real-time updates

### Layer 2: Orchestration
- agent-orchestrator: LLM-driven interview agent and background plan generator
- Conducts voice/text interviews
- Runs background cognition loops
- Resolves contradictions in evidence

### Layer 3: Core Models
- trajectory-core: Life physics engine (TypeScript/Node)
- Maintains latent state `z_t`
- Computes skill graph, alignment, gravity, mass, escape index
- Emits state change events

### Layer 4: ML Models (Python)
- skill_graph: Bayesian competency graph with message passing
- alignment: Project coherence scorer using embeddings
- gravity_mass: Constraint and inertia estimators
- life_state: State transition dynamics (`z_{t+1} = h(z_t, u_t)`)

### Layer 5: Memory & Ingestion
- vector-store: Semantic memory (embeddings of artifacts, transcripts, plans)
- artifact-ingestor: Pipelines for docs, code, calendar, etc.

### Layer 6: Future Integration
- echelon-bridge (stub): Will map embodied signals (tension, flow, drift) to life variables

Data Flow

User Interview → Agent Orchestrator → Evidence Extraction
                                    ↓
                          Trajectory Core ← Python Models
                                    ↓
                          State Update (z_t, η_t, etc.)
                                    ↓
                          Dashboard + Events

Key Concepts

### Life State `z_t`
A latent vector encoding:
- Capability profile across skills
- Alignment between projects
- Environmental constraints (gravity)
- System inertia (mass)
- Momentum and trajectory direction

Escape Index `η`

η = Thrust / (Gravity × Mass)

• η < 1: Stuck in gravity well
• η ≈ 1: At escape velocity threshold
• η > 1: Self-sustaining upward trajectory

### The Moat
When Echelon integrates, TrajectoryOS becomes the only system that can:
- Detect alignment through embodied flow states
- Measure gravity through movement tension
- Track momentum through rhythm coupling
- Validate verbal claims against physical reality

This makes the system unreplicable without years of embodied research.

Service Communication

Services communicate via:
- REST APIs for synchronous queries
- Event bus (PostgreSQL LISTEN/NOTIFY or message queue) for state changes
- Shared database for trajectory state and time-series

Future Phases

1. Phase 1 (current): Core trajectory modeling without embodied signals
2. Phase 2: Echelon integration for embodied state fusion
3. Phase 3: Multi-user, coaching, and enterprise features