The complete technical deep dive. Every strategy, every indicator, every risk layer — explained. No black boxes.
MagicMoneyMachine is built on one thesis: crypto assets with deflationary economics increase in value over time. The question isn't whether to buy — it's when to buy and how much to buy.
Every trade follows the same pattern: enter when the market is oversold and fearful, scale out at predetermined profit targets, and never panic-sell established coins. Laddered entries, Fibonacci exits, no emotional decisions.
The system doesn't predict prices. It doesn't chase pumps. It identifies statistically extreme conditions where mean reversion has a 92%+ probability of producing profit, then enters with precise position sizing and exits at predetermined levels.
Key insight: In crypto, the best time to buy is when it feels the worst. The Fear & Greed index at 10 (extreme fear) has historically preceded 85%+ of major recoveries. The system is designed to deploy capital when humans are too afraid to click "buy."
Every buy must pass through 8 sequential gates before a single dollar is deployed. If any gate fails, the trade is rejected. No overrides, no exceptions.
| Gate | What It Checks | Why It Matters |
|---|---|---|
| 1. MR Confirmation | Mean reversion brain must signal a buy (z-score < -2.0, RSI < 55) | 92% win rate gatekeeper — no entry without statistical confirmation |
| 2. Consensus | Multiple engines must agree on direction | Prevents single-engine false signals. 3+ engines agreeing = 2x position size |
| 3. Screener Rank | Coin must score well on 7-dimension analysis | Picks the best opportunity from all available coins |
| 4. Blacklist | 30-day rolling P&L per coin | Coins with consistent losses get banned automatically |
| 5. Correlation | BTC-correlated exposure capped at 50% | Prevents portfolio concentration — crypto correlations spike during crashes |
| 6. Capital | Position size, risk budget, cash reserve | Never risk more than 2% per trade. Always keep dry powder |
| 7. TA Convergence | 10 technical indicators must align (score > 0.30) | Blocks entries where TA headwinds exist (strong trend, distribution, overbought) |
| 8. Spread Check | Bid-ask spread must be below 1.5% | Prevents execution in illiquid conditions where slippage eats the edge |
ADX Regime Routing: When the mean reversion engine has no signals (market isn't oversold), the system checks if a strong trend exists (ADX ≥ 25). If so, it routes to the Trend Follower engine instead of sitting idle. Different market conditions get different tools.
The core strategy is mean reversion — the statistical tendency for prices to return to their average after extreme moves. When price drops far below its normal range, the odds of a bounce are heavily in your favor.
VWAP (Volume-Weighted Average Price) is the "true" price — where most volume traded. The z-score measures how many standard deviations the current price is from VWAP. A z-score of -2.0 means the price is 2 standard deviations below normal — a statistically rare event that tends to reverse.
Entry condition: Z-score < -2.0 AND RSI < 55. This combination means the price is both statistically extreme (z-score) and has bearish momentum that's starting to exhaust (RSI not deeply oversold means the selling pressure is fading).
Why 92% win rate? Because the system only buys at extreme deviations. In a 280-day backtest, the mean reversion engine correctly predicted a profitable bounce 92% of the time. The 8% that failed were caught by stop losses and crash DCA.
Why VWAP, not SMA? Crypto trades 24/7 with no closing price. Simple moving averages weight all candles equally, but VWAP weights by volume — giving more importance to prices where real money traded. This makes it the institutional anchor price in crypto.
Research: Quantitative Trading by Ernest Chan; Statistical Arbitrage by Andrew Pole
No single indicator is reliable alone. The TA Convergence engine combines 10 indicators into a single 0-1 score. When multiple indicators agree, confidence is high. When they disagree, the system reduces position size or blocks the entry entirely.
| Score Range | Action | What Happens |
|---|---|---|
| ≥ 0.70 | Boost | Position size increased 1.3x — multiple indicators confirm the setup |
| 0.35 – 0.70 | Neutral | Normal position size — no strong signal either way |
| 0.30 – 0.35 | Reduce | Position size cut to 0.6x — some headwinds present |
| < 0.30 | Block | Entry rejected — too many indicators disagree |
The system runs 6 independent strategy engines, each optimized for different market conditions. The Engine Promoter tracks their accuracy and promotes/demotes them based on real performance data.
| Engine | Strategy | Best Market | Status |
|---|---|---|---|
| Mean Reversion | Buy when price < VWAP - 2 std devs, RSI < 55 | Choppy / ranging | Live (92% WR) |
| Trend Follower | SMA20 > SMA50, MACD positive, volume rising | Strong uptrends | Shadow |
| Dip Buyer | 3-8% pullback in uptrend (EMA12 > EMA26) | Bull market dips | Shadow |
| Momentum Breakout | Price above 20-day high on 2x volume | Breakout rallies | Shadow |
| VSA Breakout | Smart money accumulation + rising OBV + ADX > 20 | Pre-breakout | Shadow |
| Golden Hours | MR signals boosted 2x during 03-07 UTC | Low-volume hours | Shadow |
New engines start in shadow mode: they generate signals that are logged but not executed. After 15+ resolved signals, the Engine Promoter checks their performance:
This means only proven strategies trade real money. The system continuously tests new ideas without risking capital.
Research: Evidence-Based Technical Analysis by David Aronson
The system doesn't try to pick the top. Instead, it sells in predetermined slices at Fibonacci-derived profit levels. This locks in gains progressively while keeping exposure for larger moves.
Each position is divided into 4 sell targets based on Fibonacci ratios above the entry price:
| Target | Fibonacci Level | % Above Entry | Sell Portion | Purpose |
|---|---|---|---|---|
| T1 | 0.236 | +3% | 20% | Quick profit — recover fees + lock base gain |
| T2 | 0.382 | +5% | 25% | Core profit — builds the winning track record |
| T3 | 0.618 | +8% | 30% | Momentum profit — captures the meat of the move |
| T4 | 1.000 | +13% | 25% | Full extension — let winners run to completion |
These sell orders are placed as GTC (Good-Till-Cancelled) limit orders directly on Coinbase. They execute automatically even if the bot is offline — you get maker fees (lower cost) and guaranteed fills at your target price.
Why Fibonacci? These ratios (0.236, 0.382, 0.618, 1.0) appear throughout markets as natural support/resistance levels. They're derived from the mathematical pattern where each number is the sum of the two before it. Traders worldwide watch these levels, making them self-fulfilling.
Research: Fibonacci Trading by Carolyn Boroden; Technical Analysis of the Financial Markets by John Murphy
Traditional stop-losses sell your position when it drops a certain percentage. In crypto, this is often the worst possible action — you sell at the bottom right before the recovery. MagicMoneyMachine does the opposite.
Instead of selling at -30% (locking in a loss), the system places limit BUY orders at -15%, -25%, and -35% below entry. When the price crashes, it buys more — lowering your average cost so recovery means faster profit.
| Drop Level | Top Tier (BTC, ETH) | Growth Tier (All Others) |
|---|---|---|
| -15% below entry | 50% of initial position | 30% of initial position |
| -25% below entry | 40% of initial position | 25% of initial position |
| -35% below entry | 30% of initial position | 20% of initial position |
When a crash DCA order fills: The position is merged with a weighted-average entry recalculation. All Fibonacci sell targets are repriced from the new (lower) average. The result: you need a smaller recovery to reach profit.
Over 159 days across 30 coins, crash DCA outperformed traditional stop-losses on every metric:
| Metric | Crash DCA | Stop-Loss | Winner |
|---|---|---|---|
| Total Return | +40.48% | +39.59% | DCA |
| Win Rate | 93.6% | 92.2% | DCA |
| Max Drawdown | 14.16% | 16.19% | DCA |
Why this works in crypto: Established crypto assets (BTC, ETH, SOL, etc.) have deflationary economics and growing adoption. They've recovered from every crash in history. The question isn't if they'll recover, but when. Crash DCA turns that recovery into amplified profit because your average cost is now lower.
These orders are placed as GTC limit buys on Coinbase — they execute automatically even if the bot is offline. Bot-crash insurance that actually makes money.
No strategy survives without risk management. MagicMoneyMachine uses 6 independent protection layers, each catching what the others miss:
| Layer | Mechanism | Recovery Time | What It Catches |
|---|---|---|---|
| L1 | Auto-restart (systemd / Docker) | ~10 seconds | Process crash, OOM kill |
| L2 | Watchdog health check | ~30 seconds | Frozen process, deadlock |
| L3 | Leader election (multi-instance) | 60-90 seconds | Machine failure, network partition |
| L4 | Remote commands via Supabase | Instant (manual) | Admin override when all else fails |
| L5 | Circuit breaker auto-reset | Automatic (24h) | Consecutive losses, black swan events |
| L6 | Exchange-side orders | Immediate | Total bot failure — orders live on Coinbase |
Base risk is 2% per trade. Modified by: trade confidence (0.3-2.0x), Kelly criterion from actual win rates, local intelligence (time-of-day, consecutive losses), and Fear & Greed scaling (extreme fear = deploy more).
Hard limits: Max 25% of portfolio in any single position. Max 50% in BTC-correlated assets. Always maintain cash reserve (5-20% depending on market fear level).
If the portfolio hits a drawdown threshold, the circuit breaker activates:
/resumeResearch: Mathematics of Money Management by Ralph Vince; Trade Your Way to Financial Freedom by Van Tharp
The system improves itself through 6 feedback loops that run continuously:
| Loop | What It Does | Data Source |
|---|---|---|
| Observer | Detects patterns in the event stream — golden hours, alpha coins, exit effectiveness | events.jsonl (every trade, error, decision) |
| Engine Promoter | Promotes profitable shadow engines to live, demotes underperformers | Engine signal history (WR + profit factor) |
| Dynamic Weights | Adjusts consensus engine weights based on actual win rates (not assumptions) | Resolved trade outcomes |
| Local Intelligence | Learns best trading hours, days, holding periods from your specific trade history | Trade log + balance history |
| Coin Blacklist | Automatically bans coins with negative 30-day rolling P&L | Per-coin profit tracking |
| AI CEO | Claude AI reviews on major events (3+ loss streak, drawdown >10%, F&G extremes) | Event-driven, ~$1/year cost |
The Observer: Every trade, every decision, every error writes one line to a unified event stream. The Observer reads this stream hourly and generates actionable recommendations: "BTC wins 94% at 03-07 UTC" or "Hard stop exits cost $46.71 — consider widening." The system literally learns from its own history.
The intelligence layer aggregates data from:
Every strategy in MagicMoneyMachine is grounded in published research, not marketing claims:
| Concept | Source |
|---|---|
| Mean Reversion | Quantitative Trading — Ernest Chan |
| Volume Spread Analysis | Master the Markets — Tom Williams |
| Position Sizing (Kelly) | Mathematics of Money Management — Ralph Vince |
| Risk Management | Trade Your Way to Financial Freedom — Van Tharp |
| Behavioral Finance | Thinking, Fast and Slow — Daniel Kahneman |
| Intermarket Analysis | Intermarket Analysis — John Murphy |
| Trend Following | Trend Following — Michael Covel |
| Fibonacci Methods | Fibonacci Trading — Carolyn Boroden |
| Evidence-Based TA | Evidence-Based Technical Analysis — David Aronson |
The live dashboard shows real-time signals, positions, and performance. No signup required.
Start Trading Free