🌱 Seed System Deep Dive

Deterministic Trading Strategy Generation & Optimization

What is the Seed System?

The seed system is the core of HARVEST's strategy optimization. Similar to how Minecraft uses seeds to generate identical worlds, HARVEST uses seeds to generate reproducible trading strategies.

Each unique combination of trading parameters (indicators, thresholds, multipliers) creates a deterministic "seed" that produces identical trading behavior every time it runs. This enables:

✅ Perfect Reproducibility - Same seed always produces same trades
✅ Scientific Testing - Isolate parameter effects on performance
✅ Automatic Optimization - Test billions of combinations systematically
✅ Strategy Evolution - Track which parameters work best over time
✅ Risk Control - Blacklist underperforming strategies automatically

37.6B

Total Combinations

Tracked

4 Layers

Tracking System

Complete Audit Trail

SHA-256

Hash Verification

Immutable Snapshots

Per-TF

Independent

5 Timeframes

🔧 How Seeds Are Generated

Parameter Space

Each seed is a unique combination of these trading parameters:

Category	Parameters	Combinations
Indicators	RSI period, MACD fast/slow/signal, Bollinger period/deviation, EMA periods	~500M
Thresholds	RSI overbought/oversold, MACD crossover sensitivity, BB touch points	~200M
Position Sizing	Base multiplier, ATR multipliers, timeframe-specific scaling	~100M
Risk Management	Stop loss ATR multiplier, take profit targets, trailing stops	~150M
Time Filters	Max hold time, entry cooldown, session hours	~50M

Total: 37,600,000,000+ unique combinations possible

Seed Generation Example

# Example seed configuration seed_config = { "rsi_period": 14, "rsi_oversold": 30, "rsi_overbought": 70, "macd_fast": 12, "macd_slow": 26, "macd_signal": 9, "bb_period": 20, "bb_std": 2.0, "atr_period": 14, "tp_atr_multiplier": 2.5, "sl_atr_multiplier": 1.5, "max_hold_minutes": 30 # For 1m timeframe } # This generates seed: 1234567890 # Same config always produces seed: 1234567890 # Same seed always produces same trades

📊 4-Layer Tracking System

Layer 1: Seed Registry

ml/seed_registry.json

Purpose: Master database of all tested seed configurations

Contains:

Complete parameter configurations for each seed
Aggregate statistics (total trades, win rate, P&L, max drawdown)
Test history (when tested, data range, timeframe)
Status flags (active, whitelisted, blacklisted)
Last updated timestamp

Example Entry:

{ "1234567890": { "config": { "rsi_period": 14, "rsi_oversold": 30, "tp_atr_multiplier": 2.5, ... }, "stats": { "total_trades": 150, "wins": 108, "losses": 42, "win_rate": 0.72, "total_pnl": 45.67, "max_drawdown": 8.23 }, "timeframe": "1m", "status": "whitelisted", "last_tested": "2024-12-18T08:30:00" } }

Layer 2: Seed Snapshots

ml/seed_snapshots.json

Purpose: Immutable configuration verification with cryptographic hashing

Contains:

SHA-256 hash of configuration (prevents tampering)
Configuration snapshot at time of creation
Creation timestamp (ISO 8601)
Reproducibility verification data

Why SHA-256?

Ensures configuration hasn't been modified
Verifies same config produces same seed
Detects any parameter drift or corruption
Enables blockchain integration (future)

Example Entry:

{ "1234567890": { "sha256_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "snapshot": { "rsi_period": 14, ... }, "created_at": "2024-12-18T08:30:00Z", "verified": true } }

Layer 3: Seed Catalog

ml/seed_catalog.json

Purpose: Detailed trade-by-trade records with searchable metadata

Contains:

Every individual trade with entry/exit details
Daily performance breakdown
Searchable fields (seed, timeframe, date, asset, profit)
Position sizing and risk metrics per trade

Use Cases:

Analyze specific seed performance over time
Identify best/worst trading days
Debug strategy behavior on specific dates
Compare seeds on identical data

Example Entry:

{ "trades": [ { "seed": "1234567890", "timeframe": "1m", "asset": "ETH", "date": "2024-12-18", "entry_time": "08:30:00", "exit_time": "08:45:00", "entry_price": 2250.50, "exit_price": 2258.30, "position_size": 3.00, "profit": 0.10, "win": true, "hold_duration_minutes": 15 }, ... ] }

Layer 4: Performance Tracker

ml/seed_performance_tracker.json + separate whitelist/blacklist files

Purpose: Automatic strategy optimization based on performance

Contains:

Real-time performance monitoring per seed
Automatic whitelist/blacklist decisions
Threshold tracking (win rate, P&L, trade count)
Separate files for whitelist and blacklist

Whitelist Criteria:

✅ Win rate ≥ 70%
✅ Total P&L > 0
✅ Minimum 15 trades completed
✅ Max drawdown < 15%

Blacklist Criteria:

❌ Win rate < 55%
❌ Total P&L < 0 (negative)
❌ Max drawdown > 20%
❌ Consistent underperformance

🧪 Testing Workflows

1. Test All Whitelisted Seeds

# Generate test file with all whitelisted seeds for 1m timeframe python ml/seed_tester.py test-all 1m # Output: ml/test_all_whitelisted_1m.json # Contains all seeds with WR ≥ 70%, positive P&L # Run backtest with all seeds python backtest_90_complete.py --test-seeds-file ml/test_all_whitelisted_1m.json # Results show which seeds perform best on recent data

2. Test Top 10 Performers

# Generate test file with top 10 seeds for 15m timeframe python ml/seed_tester.py test-top10 15m # Output: ml/test_top10_15m.json # Contains 10 seeds with highest combined WR × P&L score # Run backtest python backtest_90_complete.py --test-seeds-file ml/test_top10_15m.json # Identify the single best performer

3. Deploy Best Seed to Production

# After testing, overwrite BASE_STRATEGY with best performer python ml/seed_tester.py overwrite 15m --use-best # This updates core/strategy_config.py: # BASE_STRATEGY["15m"] = {best_seed_config} # Now live trading uses the optimized seed for 15m

4. Check System Status

# View current BASE_STRATEGY configuration python ml/seed_tester.py status # Shows active seed for each timeframe: # 1m: Seed 1234567890 (WR: 75%, P&L: +12.34) # 5m: Seed 9876543210 (WR: 72%, P&L: +8.91) # 15m: Seed 5555555555 (WR: 68%, P&L: +15.67) # 1h: Seed 3333333333 (WR: 70%, P&L: +22.45) # 4h: Seed 7777777777 (WR: 65%, P&L: +30.12)

5. Reset to Defaults

# Reset specific timeframe to default BASE_STRATEGY python ml/seed_tester.py reset --timeframe 1m # Reset all timeframes python ml/seed_tester.py reset

⚡ Per-Timeframe Independence

Each timeframe has its own independent seed and optimization.

This means:

1m timeframe uses seed optimized for ultra-fast scalping (30 min hold max)
5m timeframe uses seed optimized for fast trades (1 hour hold max)
15m timeframe uses seed optimized for short-term (3 hour hold max)
1h timeframe uses seed optimized for medium-term (12 hour hold max)
4h timeframe uses seed optimized for long-term (24 hour hold max)

Why Independent?

Each timeframe has unique market dynamics
Optimal parameters differ (1m needs faster indicators than 4h)
Risk profiles vary (1m = frequent small trades, 4h = rare large swings)
Testing one timeframe doesn't affect others

🔬 Backtest Integration

Test Specific Seed

# Test seed 1234567890 on 90 days of data python backtest_90_complete.py --seed 1234567890 # Results show: # - Total trades # - Win rate # - P&L per timeframe # - Drawdown # - Position sizing stats

Test Multiple Seeds

# Create custom test file cat > ml/custom_seeds.json << EOF { "seeds": [ {"seed": "1111111111", "timeframe": "1m"}, {"seed": "2222222222", "timeframe": "5m"}, {"seed": "3333333333", "timeframe": "15m"} ] } EOF # Run batch test python backtest_90_complete.py --test-seeds-file ml/custom_seeds.json

Compare Seed Performance

After running multiple tests, compare results:

Win rate per timeframe
Total P&L
Max drawdown
Average trade duration
Position sizing efficiency

✅ Best Practices

1. Regular Optimization

Test seeds weekly on fresh data
Update BASE_STRATEGY if better performers found
Track performance after deployment

2. Validation Before Deployment

Always test on 90 days of data minimum
Require 100+ trades for statistical significance
Verify consistent performance across different market conditions

3. Monitor Degradation

Seeds can degrade over time as market changes
Watch for declining win rates or increasing drawdowns
Re-test periodically even if currently whitelisted

4. Timeframe-Specific Testing

Test each timeframe independently
Don't use 1m seeds for 4h trading (different dynamics)
Optimize per-timeframe for best results

📊 Dashboard Integration

The dashboard displays seed information in real-time:

Seed Status Panel

Shows active seed for each timeframe
Displays seed performance (WR, P&L)
Indicates whitelist/blacklist status
Updates as trades execute

Performance Tracking

Live win rate per seed
Running P&L totals
Trade counts
Position sizing effectiveness

Press 'S' in dashboard to browse all seeds, whitelist, and blacklist!

📝 Summary

The seed system enables scientific strategy optimization at scale.

Key Advantages:

✅ 37.6 billion combinations tracked systematically
✅ Perfect reproducibility for every strategy
✅ 4-layer tracking with SHA-256 verification
✅ Automatic whitelist/blacklist optimization
✅ Per-timeframe independence for specialized strategies
✅ Complete audit trail of all tests and trades
✅ Easy deployment of proven strategies to production

This is how HARVEST achieves 70%+ win rates consistently - by scientifically testing and deploying only the best-performing strategies.

← Back to Home User Manual Mathematics