The Crypto ML Edge Engine uses academically proven, statistically validated strategies -- not black-box ML predictions.
Every pick comes with a full audit trail showing exactly why it was generated, what indicators triggered, and the academic source backing the strategy.
The system runs autonomously via GitHub Actions, scanning markets every 30 minutes and tracking all open positions against live prices.
Goes long on extreme negative funding. Blocked: Binance Futures API geo-restricted from CI.
BLOCKED
Crypto
71.0%
Feb 18
Feb 18
Binance perp funding research
Hover over dates for details. Green dates = updated in last 48h.
Strategies labeled PROVEN have statistically significant backtests (p < 0.05). NEW strategies are in forward-testing.
System Operations
Parameter
Value
Details
Scan Frequency
Every 30 min
GitHub Actions cron schedule. Checks for new signals + tracks open positions.
Capital Base
$10,000
Simulated starting capital. Position sizes are 2-5% per trade based on confidence.
Max Concurrent Picks
10
Prevents over-concentration. New picks queued if all slots full.
Position Sizing
2-5% of capital
Scaled by confidence: 60% conf = 2%, 75%+ conf = 4.4%. Hard cap at 5%.
Max Hold Time
10 bars
Auto-closes positions after 10 daily bars. Mean reversion should resolve in 2-5 days — if it hasn't by day 10, the thesis is wrong.
Confidence Threshold
60%
Picks below 60% are filtered out. Only high-conviction signals are published.
Risk/Reward
1.2-1.5:1
TP is 1.2x ATR (crypto) or 1.5x ATR (ETF). Tight targets match mean reversion reality — capture the bounce, not a trend reversal.
Falling Knife Filter
20% max below SMA
Rejects signals where price is >20% below 200 SMA. Connors RSI-2 was proven on temporary dips in uptrends, not structural bear markets.
Price Tracking
Binance + CoinGecko
Tries Binance API first, falls back to CoinGecko for CI environments where Binance is geo-blocked (451).
Data Refresh
5 min (dashboard)
Dashboard auto-fetches latest active_picks.json every 5 minutes from GitHub raw.
Tier Classification
Tier
Criteria
Position Size
HIGH
Confidence ≥ 75% AND Risk/Reward ≥ 1.5:1
~4.4% ($440)
MEDIUM
Confidence ≥ 65%
~3.6% ($356)
WATCH
Confidence 60-65%
~2.0% ($200)
Statistical Validation
Backtest
Win Rate
p-value
Sharpe
Status
Connors RSI-2 on SPY (2000-2025)
75.7%
6e-06
4.84
PROVEN
Connors RSI-2 on QQQ (2000-2025)
75.3%
8e-06
6.55
PROVEN
VIX Spike Reversal (2008-2025)
72.0%
0.022
6.20
PROVEN
Connors RSI-2 on BTC (2017-2025)
62.5%
0.009
2.35
PROVEN
Funding Rate Carry on DOGE
71.0%
~0.042
8.19
MARGINAL
*Strategy Generations
Gen 1 (Feb 18-23): Connors RSI-2, Funding Rate Carry, VIX/Fear Reversal. These are the proven foundation strategies with backtested win rates of 71-76%.
Gen 2 (Feb 24+): Fib Trend Pullback (3-layer confluence) + expanded asset universe (SLV, VDE, COPX). These are newer strategies in forward-testing.
ML Pilots (Feb 24+): TA Ensemble (22-feature LightGBM), News Sentiment (headline scraping + contrarian), Multi-Asset Momentum (cross-asset signals). Training and calibrating.
Killed: double_top_bottom_detector was disabled after -$14,088 loss (5.6% WR on 18 trades).