Executive Dashboard
Stocks
Crypto
Forex
Sports Betting
Penny Stocks
Meme Coins
ML & Learning
World-Class Gaps
Forward-Facing vs Backtested Performance • Machine Learning Progress • World-Class Gap Analysis
| Asset Class | Backtested Sharpe | Backtested Grade | Backtested Trades | Forward WR | Forward Trades | Forward PnL | Overlap | Status |
|---|
Our trading intelligence platform operates across three independent layers:
1. Daily Picks Engine (7 algorithms) — Generates stock picks stored in stock_picks, miracle_picks2, miracle_picks3. Algorithms: Cursor Genius, ETF Masters, Sector Momentum, Sector Rotation, Blue Chip Growth, Technical Momentum, Composite Rating. These are backtested over 2 years but not running in live paper trading.
2. Live Signal Engine (20 algorithms) — Real-time signals via live_signals.php. Includes Ichimoku Cloud, StochRSI Crossover, MACD Divergence, Bollinger Breakout, RSI Reversal, Alpha Predator, Challenger Bot, Consensus, and 12 others. These run in live paper trading but haven't completed full 2-year backtests.
3. Goldmine Tracker (8 systems) — Unified outcome verification across all prediction systems. Archives picks from Consolidated, Live Signal, Edge, Meme, Sports, Horizon, Top Picks, and Penny systems. Checks TP/SL hits against real market prices. View Dashboard →
| # | Algorithm | Sharpe | Win Rate | PF | Total PnL | Trades | Max DD | Grade |
|---|
| # | Algorithm | Sharpe | Win Rate | PF | Total PnL | Trades | Max DD | Grade |
|---|
| # | Algorithm | Sharpe | Win Rate | PF | Total PnL | Trades | Max DD | Grade |
|---|
Our system uses a closed-loop self-optimization pipeline:
Step 1: Signal Generation — 20 algorithms generate BUY/SELL/SHORT signals using technical indicators (RSI, MACD, Ichimoku, Bollinger, etc.). Each signal is tagged with the algorithm name and default parameters (TP%, SL%, max_hold).
Step 2: Trade Execution — Paper trading engine (live_trade.php) opens positions based on signals. Position sizing: 5% of capital (~$500). Max 10 concurrent positions. Regime gating blocks signals that disagree with HMM-detected market state.
Step 3: Outcome Tracking — Each trade is tracked until exit (TP hit, SL hit, or max hold reached). Exit reason, hold time, and P&L are recorded in lm_trades table.
Step 4: Grid Search Optimization — After accumulating 20+ closed trades per algorithm-asset pair, the learning system (learning.php) runs a grid search over TP (1-10%), SL (0.5-5%), and max_hold (4-72h) combinations to find parameters that maximize Sharpe ratio on historical trades.
Step 5: Parameter Evolution — Learned parameters replace defaults for future signals. Performance is tracked as "Learned vs Original" on the L vs O Dashboard.
Step 6: Goldmine Verification — The Goldmine Tracker independently verifies all predictions against real market prices across 8 systems, surfacing algorithms that perform well in forward conditions.
| Algorithm | Default TP | Learned TP | Change | Default SL | Learned SL | Change | Default Hold | Learned Hold | Change | Result |
|---|---|---|---|---|---|---|---|---|---|---|
| Ichimoku Cloud | 2% | 3.4% | +70% | 1% | 2% | +100% | 16h | 16h | 0% | 80% WR, +$44.97 |
| StochRSI Crossover | 2% | 3.1% | +55% | 1% | 2% | +100% | 12h | 12h | 0% | 75% WR, +$4.25 |
| Consensus | 3% | 3% | 0% | 2% | 2% | 0% | 24h | 12h | -50% | 0% WR, -$3.28 |
| RSI Reversal | 2% | 2% | 0% | 1% | 1% | 0% | 12h | 6h | -50% | 0% WR (1 trade) |
learned_params API returns empty despite evidence of parameter evolution in by_algorithm data. The system IS learning (TP/SL adjustments visible), but the stored params endpoint hasn’t been populated yet.lm_learned_params table from current performance data. ETA: Next code update.
Sound practices: Walk-forward optimization (no peeking at future data), commission modeling (Questrade tiered), slippage (0.1%), 2-day embargo after pick date, regime detection, grid search with Sharpe maximization objective.
Known risks: Small sample size (13 live trades), potential overfitting of learned parameters to recent market regime, no out-of-sample validation set in grid search, limited asset universe (14 crypto, 12 stocks, 10 forex pairs).
Missing (industry standard): Cross-validation with time-series splits, Monte Carlo simulation of parameter sensitivity, correlation analysis between positions, factor exposure analysis (beta, sector, momentum), transaction cost optimization.
Renaissance Technologies (Medallion Fund) — Sharpe ~3.0–6.0+. Annual return ~66% gross / ~39% net. Max drawdown typically <5%. Uses 10,000+ trading signals, microsecond execution, statistical arbitrage and mean reversion on short time horizons. $10B+ AUM. Widely considered the most successful quant fund in history.
Two Sigma — Sharpe ~1.5–2.5. $60B+ AUM. ML/AI-driven with massive data infrastructure. Medium-frequency strategies. PhD-level research teams. Known for systematic, data-driven approach.
Citadel (Wellington) — Sharpe ~1.5–3.0. Multi-strategy (fundamental + quantitative). Market-making arm (Citadel Securities) has near-100% daily win rate. Technology and execution speed advantages.
D.E. Shaw — Sharpe ~1.5–2.5. Hybrid quantitative + fundamental. Computational modeling expertise. One of the original quant funds.
AQR Capital — Sharpe ~1.2. Factor-based investing (value, momentum, carry, defensive). Academic research-driven. Transparent methodology.
Price Data: Yahoo Finance (stocks, daily OHLCV), Finnhub (stocks, real-time quotes), TwelveData (forex, 10 pairs), FreeCryptoAPI (crypto, 14 pairs). Limitation: daily bars for backtesting miss intraday dynamics. Real-time feeds have 1-second minimum resolution.
Backtest Period: Feb 2024 – Feb 2026 (2 years). Limitation: single market cycle. Does not include 2020 COVID crash or 2022 bear market. May overfit to recent bull/sideways regime.
Commission Model: Questrade tiered ($4.95–$9.95 per trade) + 0.1% slippage. Limitation: high commission impact on small positions ($500). World-class firms use IBKR ($0.005/share) or direct market access ($0.001/share). Commission drag is our single biggest performance killer.
Position Sizing: Fixed 5% of capital ($500 per trade). Limitation: not risk-adjusted. World-class firms use volatility-adjusted sizing (risk parity), Kelly criterion, or EGARCH-based sizers.
Regime Detection: Hidden Markov Model (HMM) with 3 states (bull/sideways/bear). Limitation: HMM is fitted on recent data, may lag regime transitions by 1–3 days. Not validated against known regime change dates.
Signal Universe: 20 algorithms across 36 assets (12 stocks, 14 crypto, 10 forex). Limitation: Renaissance uses 10,000+ signals across thousands of assets. Our signal diversity is ~500x less.
Dashboard: Sports Betting Dashboard — Real-time value bet finder with line shopping across 20+ bookmakers
Data Source: The Odds API (free tier, 500 credits/month). Covers 8 sports: NHL, NBA, NFL, MLB, CFL, MLS, NCAAF, NCAAB. 20+ bookmakers (FanDuel, DraftKings, bet365, Pinnacle, etc.)
APIs: sports_odds.php (odds cache), sports_picks.php (value bets + line shopping), sports_bets.php (paper betting), sports_ml.php (ML scoring)
Automation: GitHub Actions runs 5x daily (10am, 1pm, 4pm, 7pm, 10pm EST). Fetches odds, runs Kimi scrapers (stats/injuries/schedule), scores with ML, auto-places top value bets (EV ≥ 3%), auto-settles completed bets.
Goldmine: Tracked in Goldmine Dashboard as sports_bets + sports_picks systems
Step 1: Odds Collection — Every ~5 hours, fetch real-time odds from 20+ sportsbooks via The Odds API. Cache in lm_sports_odds table.
Step 2: Value Bet Analysis — Compare each bookmaker’s odds against the sharp market (Pinnacle). When a book offers odds significantly above consensus, flag as a value bet with expected value (EV%).
Step 3: ML Scoring — Score each value bet using features: historical Closing Line Value (CLV), team stats (NBA/NHL/NFL/MLB), injuries, schedule, and market consensus.
Step 4: Auto-Placement — Top value bets (EV ≥ 3.0%) are auto-placed as paper bets. Position sizing: quarter-Kelly criterion (max 5% of bankroll per bet, 20 max concurrent positions).
Step 5: Settlement — After games complete, auto-settle using final scores. Track P&L, ROI, and win rate by sport and algorithm.
Dashboard: Penny Stock Finder — Multi-tab dashboard: Finder, Picks, History, Performance
Data Source: Yahoo Finance screener API (crumb-authenticated). Filters: exchange-listed only (blocks OTC/Pink Sheets), price $0.01–$5.00, min volume 100K.
APIs: penny_stocks.php (Yahoo screener proxy, 30min cache), penny_stock_picks.php (daily picks + tracking)
Automation: GitHub Actions weekdays 7:00 AM EST. Python engine (scripts/penny_stock_picks.py) scores and ranks penny stocks using 8-factor composite score. Tracking job updates prices and checks TP/SL.
Goldmine: Tracked in Goldmine Dashboard as penny_stocks system
Each penny stock is scored across 8 dimensions to produce a composite score:
Dashboard: Meme Coin Scanner — Winners leaderboard, scan history, educational resources
Data Source: Crypto.com Exchange API (public, no auth) + CoinGecko fallback. Real-time OHLCV data for Tier 1 + discovered Tier 2 meme coins.
APIs: meme_scanner.php (scan + score + winners), meme_ml_engine.php (adaptive ML model)
Automation: GitHub Actions scans every 10 minutes (memes move fast). Resolves signals every 3 hours (2h outcome window). Daily ML retraining check at 6 AM UTC.
Goldmine: Tracked in Goldmine Dashboard as meme_scanner system
DOGE, SHIB, PEPE, FLOKI, BONK, WIF, TURBO, NEIRO
Most liquid meme coins with established markets. Scanned every 10 minutes regardless of volume conditions.
PNUT, GOAT, ACT, CHILLGUY, SPX, and 50+ more
New meme coins are automatically discovered when they appear on Crypto.com with matching keywords. Volume filters apply.
Each coin is scored across 7 technical/volume features. Tiers: STRONG (80+) BUY (65-80) LEAN (50-65)
meme_ml_engine.php) periodically retrains on historical signal outcomes. It redistributes feature weights based on which indicators actually predicted winners. Auto-retrain triggers after 50 new signals or 7 days since last training.