Skip to main content

Mercury Instance Architecture: Strategic Refinement of W/R/ABH System

· 3 min read
Max Kaido
Max Kaido
Architect

TL;DR: The Strategic Breakthrough​

After deep architectural analysis, we've discovered the true competitive advantage of Mercury's instance system:

  • W Instance: Live trading with conservative risk management
  • R Instance: Demo trading (Bybit testnet) mirroring W constraints
  • ABH Instance: Shadow portfolio as unrestricted market intelligence engine

Key Insight: Shadow portfolio's value isn't simulation—it's complete data capture for strategy optimization that demo trading cannot provide.

Current Mercury Instance System​

W Instance (Live Trading)​

MERCURY_INSTANCE = W;
// Real Bybit API (testnet: false)
// Conservative risk management: strict RR requirements
// Selective execution: topL from topK based on risk constraints
// Time slot: 36-72 minutes in 3-hour blocks
// Database: postgres-mercury-w (port 5436)

R Instance (Shadow Production)​

MERCURY_INSTANCE = R;
// No external API - pure internal simulation
// Same MAIN variant as W instance
// Time slot: 0-36 minutes in 3-hour blocks
// Database: postgres-mercury-r (port 5435)

ABH Instance (Experimental Consolidated)​

MERCURY_INSTANCE = ABH;
// Runs experimental variants A, B, H
// Time-based variant selection within 72-180 minute window
// 24 tournaments per cycle: 8 base × 3 variants
// Database: postgres-mercury-abh (port 5439)

The Strategic Refinement: W/R/ABH Redesign​

Proposed New Architecture​

W Instance (Live Trading)​

  • API: Real Bybit production API
  • Risk Management: Conservative - strict RR thresholds
  • Position Creation: Only topL positions meeting risk criteria
  • Purpose: Actual money trading with proven strategies

R Instance (Demo Trading)​

  • API: Bybit demo/testnet API (api-demo.bybit.com)
  • Risk Management: Mirror W instance constraints exactly
  • Position Creation: Same topL logic as W for validation
  • Purpose: Risk-free validation of live trading strategies

ABH Instance (Market Intelligence Engine)​

  • API: No external API - pure shadow simulation
  • Risk Management: UNRESTRICTED - ignore RR requirements
  • Position Creation: ALL topK positions without filtering
  • Purpose: Complete market intelligence for strategy optimization

The Game-Changing Insight: Why Shadow > Demo for Intelligence​

What Demo Trading Cannot Provide​

// Demo Trading (R instance) - CONSTRAINED DATA
const positions = topK.filter((signal) => {
  return signal.riskReward >= MIN_RR_RATIO; // ❌ FILTERS OUT VALUABLE DATA
});
await createDemoPositions(positions);

What Shadow Portfolio Captures​

// Shadow Portfolio (ABH instance) - COMPLETE DATA
const positions = topK.map((signal) => {
  return createShadowPosition(signal); // ✅ CAPTURES EVERYTHING
});
// No filtering = complete market intelligence

Critical Research Questions Only Shadow Can Answer​

  1. Optimal RR Discovery

    • "What if we lowered RR threshold from 2.0 to 1.5?"
    • "Which RR threshold maximizes Sharpe ratio historically?"

  2. Dynamic topL Optimization

    • "Should we trade top 3 or top 7 positions?"
    • "Do positions 6-10 consistently lose money?"
    • "What's the optimal topL for each tournament type?"

  3. Market Regime Analysis

    • "Do all topK perform well during volatility breakouts?"
    • "Which positions fail during range-bound markets?"
    • "Should we skip certain positions in specific regimes?"

  4. Strategy Refinement

    • "Are momentum signals better than mean reversion in positions 4-6?"
    • "What's the performance distribution across all tournament winners?"

The Analytics Pipeline​

// Phase 1: Shadow Portfolio - Unrestricted Data Collection
shadowPortfolio.createAllPositions(topK); // Capture everything

// Phase 2: Post-hoc Analysis
const strategyAnalysis = {
  rrThresholds: [1.2, 1.5, 2.0, 2.5, 3.0],
  topLCounts: [1, 3, 5, 7, 10],
  marketRegimes: ['trending', 'ranging', 'volatile'],
  timeframes: ['1h', '4h', '1d'],
};

// Phase 3: Optimization Insights
const optimalStrategy = analyzeHistoricalPerformance(
  shadowData,
  strategyAnalysis,
);

// Result: "Use RR≥1.5, topL=5, skip positions 7+ in ranging markets"

Implementation Strategy​

Phase 1: Instance Purpose Clarification​

// W Instance: Live + Conservative
if (instance === MercuryInstance.W) {
  const filteredPositions = topK.filter(meetsStrictRiskCriteria);
  await bybitService.createLivePositions(filteredPositions);
}

// R Instance: Demo + Conservative (mirrors W exactly)
if (instance === MercuryInstance.R) {
  const filteredPositions = topK.filter(meetsStrictRiskCriteria);
  await bybitDemoService.createDemoPositions(filteredPositions);
}

// ABH Instance: Shadow + Unrestricted Intelligence
if (instance === MercuryInstance.ABH) {
  const allPositions = topK; // No filtering!
  await shadowPortfolioService.createShadowPositions(allPositions);
}

Phase 2: Bybit Demo Integration​

// R Instance Configuration
const bybitDemoClient = new RestClientV5({
  key: demoApiKey,
  secret: demoApiSecret,
  testnet: true, // 👈 Demo API
  baseUrl: 'https://api-demo.bybit.com',
});

Phase 3: Analytics Dashboard Enhancement​

  • Historical RR threshold performance analysis
  • Dynamic topL optimization recommendations
  • Market regime correlation studies
  • Strategy parameter backtesting interface

Competitive Advantages​

1. Complete Market Intelligence​

  • No data loss from filtering
  • Retroactive strategy optimization
  • Comprehensive performance analytics

2. Risk-Free Strategy Validation​

  • R instance mirrors W exactly with demo API
  • Zero risk validation of live strategies
  • Perfect testing environment

3. Systematic Optimization​

  • Data-driven strategy refinement
  • Evidence-based parameter tuning
  • Continuous improvement loop

4. Hardware Isolation Safety​

  • Only W instance has live API keys
  • Impossible accidental live trading from experiments
  • Clean separation of concerns

The Strategic Value Proposition​

This architecture transforms Mercury from "trading bot with experiments" into "adaptive trading intelligence system":

  1. W Instance: Executes proven strategies with real money
  2. R Instance: Validates strategies risk-free before live deployment
  3. ABH Instance: Continuously discovers better strategies through complete data analysis

Key Insight: Shadow portfolio isn't just simulation—it's our competitive moat for strategy optimization that no demo trading system can replicate.

Next Steps​

  1. Refactor R instance to use Bybit demo API instead of shadow simulation
  2. Enhance ABH analytics to capture all topK positions without filtering
  3. Build optimization dashboard for historical strategy analysis
  4. Implement gradual rollout of optimized strategies from ABH to R to W

Bottom Line: We've discovered that our shadow portfolio system provides irreplaceable market intelligence that demo trading cannot match. This isn't just architecture—it's our strategic advantage for systematic trading optimization.