How atul2501/reserch_model works
This is a paper-first crypto strategy research and trading platform: a worker turns confirmed market candles into per-agent decisions, applies deterministic risk controls, simulates execution, and records outcomes. A separate scheduled research process evaluates and stress-tests strategies, evolves candidates, and gates promotion using a protected out-of-sample lockbox and champion/challenger governance. The API exposes operational and research views. The README documents the product workflow; sampled source confirms key risk, feature, backtest, fitness, and validation behavior, while some wiring is represented only by that documentation.
Main entry points: pipeline.py Evolution Pipeline · cycle.py Trading Worker · decision_loop.py Agent Decisions · router.py Execution Router
Components 24
Market and Trading
- Trading Worker
backend/app/worker/cycle.py - Market Data
backend/app/market/market_data_service.py - Features and Regime
backend/app/market/feature_engine.py - AI Council
backend/app/council/service.py - Agent Decisions
backend/app/agents/decision_loop.py - Strategy DNA
backend/app/strategies/engine.py - Risk Engine
backend/app/risk/risk_engine.py - Execution Router
backend/app/execution/router.py - Paper Execution
backend/app/execution/paper_adapter.py - Shadow Execution
backend/app/execution/shadow_adapter.py - Position Management
backend/app/agents/position_manager.py
Research and Evolution
- Evolution Pipeline
backend/app/research/pipeline.py - Backtesting
backend/app/backtesting/engine.py - Regime Validation
backend/app/backtesting/regime_validation_engine.py - Fitness Scoring
backend/app/analytics/fitness_engine.py - Breeding and Promotion
backend/app/evolution - OOS Lockbox
backend/app/research/lockbox.py
Analytics and Governance
- PnL Accounting
backend/app/analytics/pnl_engine.py - Application Database
backend/app/core/database.py - Performance Analytics
backend/app/analytics
API and Operations
- API Application
backend/app/main.py - API Routes
backend/app/api/routes
Other components
- Hyperliquid
- Operator and Viewer
Connections 24
- Hyperliquid → Market Data: streams market data
- Market Data → Trading Worker: feeds closed candles
- Trading Worker → Features and Regime: builds context
- Trading Worker → AI Council: requests context
- Trading Worker → Agent Decisions: dispatches candle
- Agent Decisions → Strategy DNA: evaluates DNA
- Agent Decisions → Risk Engine: checks proposed trade
- Agent Decisions → Execution Router: routes approved trade
- Execution Router → Paper Execution: dispatches paper
- Execution Router → Shadow Execution: dispatches shadow
- Trading Worker → Position Management: manages open positions
- Position Management → Execution Router: routes exits
- Position Management → PnL Accounting: computes realized PnL
- Trading Worker → Application Database: persists cycle state
- Evolution Pipeline → Backtesting: validates strategies
- Evolution Pipeline → Regime Validation: checks regime robustness
- Evolution Pipeline → Fitness Scoring: scores agents
- Evolution Pipeline → Breeding and Promotion: selects and breeds
- Evolution Pipeline → OOS Lockbox: gates OOS use
- Evolution Pipeline → Application Database: reads and records research
- Performance Analytics → Application Database: reads trade records
- API Application → API Routes: mounts routes
- Operator and Viewer → API Application: requests API
- API Routes → Application Database: reads product state