Trading

Trading Skill — Complete Reference

Purpose

Comprehensive trading knowledge base covering fundamentals, technicals, strategies, backtesting, and risk management. Use when building trading apps, evaluating strategies, or making trading decisions.

1. TRADING STYLES

Scalping

• Hold: seconds to minutes
• Goal: profit from tiny price movements
• Pros: many opportunities, reduced exposure to big moves
• Cons: high transaction costs, stressful, tiny profit per trade
• Best for: highly liquid markets with tight spreads

Day Trading

• Hold: minutes to hours (close all by market close)
• Goal: profit from intraday price movements
• Pros: no overnight risk, high profit potential per trade
• Cons: high risk, emotional pressure, costs add up
• Best for: volatile stocks with clear intraday patterns

Swing Trading

• Hold: days to months
• Goal: catch short-to-intermediate moves
• Pros: lower costs, more analysis time, less stressful
• Cons: overnight risk, may miss long-term moves
• Best for: trending markets with pullbacks

Position Trading

• Hold: months to years
• Goal: profit from major long-term trends
• Pros: lowest costs, highest profit potential, most flexibility
• Cons: capital tied up, exposed to macro events
• Best for: fundamentally strong companies in sector uptrends

2. FUNDAMENTAL ANALYSIS

Key Metrics

P/E Ratio (Price-to-Earnings)

• Formula: Share Price / Earnings Per Share
• S&P 500 average: ~30 (as of late 2025)
• Low (<15): potentially undervalued or troubled
• High (>30): potentially overvalued or high growth expected
• Compare WITHIN same industry only
• Forward P/E uses projected earnings; Trailing P/E uses last 12 months

P/B Ratio (Price-to-Book)

• Formula: Share Price / Book Value Per Share
• Book Value = total assets - intangible assets - liabilities
• P/B < 1 = trading below asset value (potential bargain)
• Most useful for capital-heavy industries (banks, manufacturing)

Debt-to-Equity Ratio

• Formula: Total Liabilities / Shareholders' Equity
• High = heavy debt reliance (risky in downturns)
• Compare within industry — some sectors carry more debt naturally

Revenue Growth Rate

• Year-over-year revenue increase
• Consistent > spiky
• Accelerating growth rate = strongest signal

Free Cash Flow (FCF)

• Cash generated after capital expenditures
• Positive FCF = real cash generation, not accounting tricks
• FCF Yield = FCF / Market Cap (higher = better value)

EPS Growth (Earnings Per Share)

• Consistent EPS growth over 3-5 years = strong signal
• Check quality: operations vs one-time events

Return on Equity (ROE)

• Formula: Net Income / Shareholders' Equity
• ROE > 15% = generally good management efficiency

Dividend Yield & Payout Ratio

• Yield = Annual Dividend / Share Price
• Payout = Dividends / Net Income
• Payout > 80% may be unsustainable

3. TECHNICAL INDICATORS

RSI (Relative Strength Index)

• Formula: RSI = 100 - (100 / (1 + (Avg Gain / Avg Loss))) over 14 periods
• Scale: 0-100
• >70 = Overbought (potential sell)
• <30 = Oversold (potential buy)
• Strengths: Simple, effective in ranging markets
• Weaknesses: Stays >70 for extended periods in strong uptrends — don't auto-sell
• RSI Divergence: Price makes new high but RSI makes lower high = bearish (momentum weakening)
• Swing Rejection: RSI crosses 30 upward → dips but stays above 30 → breaks prior high = bullish entry
• Best in: Ranging/sideways markets. Combine with ADX to filter.

MACD (Moving Average Convergence Divergence)

• MACD Line = 12-period EMA - 26-period EMA
• Signal Line = 9-period EMA of MACD Line
• Histogram = MACD Line - Signal Line
• Buy: MACD crosses ABOVE signal line
• Sell: MACD crosses BELOW signal line
• Divergence: MACD rising while price falling = potential reversal
• Strengths: Good trend-following indicator
• Weaknesses: Lagging, many false positives in sideways markets
• Best combined with ADX — only trust MACD signals when ADX > 25 (confirming trend)

Bollinger Bands

• Middle Band = 20-day SMA
• Upper Band = SMA + 2 standard deviations
• Lower Band = SMA - 2 standard deviations
• 95% of price action stays within bands
• Squeeze: Bands narrow → low volatility → breakout imminent (direction unknown)
• Bollinger Bounce: Price off lower band toward middle = buy; off upper toward middle = sell
• Breakout: Price outside bands with volume = trend continuation
• Strengths: Visual, adapts to volatility
• Weaknesses: Secondary indicator — always confirm with RSI/MACD
• Created by John Bollinger in the 1980s

Moving Averages

• SMA (Simple): Equal weight all periods
• EMA (Exponential): More weight on recent prices (faster reaction)
• Key periods: 5, 9, 20, 50, 100, 200 day
• Golden Cross: 50-day crosses ABOVE 200-day = strong bullish
• Death Cross: 50-day crosses BELOW 200-day = strong bearish
• MAs act as dynamic support/resistance levels
• Use longer MAs for volatile stocks to avoid false signals

Volume

• Confirms price movements
• Price up + high volume = strong bullish
• Price up + low volume = weak rally, likely to reverse
• Price down + high volume = strong selling pressure
• Price down + low volume = lack of selling conviction
• 60-80% of daily volume is algorithmic

Ichimoku Cloud (Ichimoku Kinko Hyo)

• 5 Components:
  • Tenkan-sen (Conversion Line): (9-period high + 9-period low) / 2
  • Kijun-sen (Base Line): (26-period high + 26-period low) / 2
  • Senkou Span A: (Tenkan + Kijun) / 2, plotted 26 periods ahead
  • Senkou Span B: (52-period high + 52-period low) / 2, plotted 26 periods ahead
  • Chikou Span: Current close plotted 26 periods back
• The Cloud (Kumo): Area between Span A and Span B = support/resistance zone
• Buy: Price above cloud, Tenkan crosses above Kijun
• Sell: Price below cloud, Tenkan crosses below Kijun
• Strengths: All-in-one indicator (trend, momentum, support/resistance)
• Weaknesses: Complex visually, lagging in fast markets
• Best in: Trending markets. Avoid in ranging/sideways.
• Michael Automates created an AI-improved Ichimoku strategy that performs well

Super Trend

• Formula: Based on ATR (Average True Range) and a multiplier
• Upper Band = (High + Low) / 2 + (Multiplier × ATR)
• Lower Band = (High + Low) / 2 - (Multiplier × ATR)
• Buy: Price crosses above Super Trend line
• Sell: Price crosses below Super Trend line
• Strengths: Simple, good trend-following, clear signals
• Weaknesses: Late entries, bad in ranging markets
• AI iteration took Super Trend from 44% to 3,605% P&L in Michael Automates' testing

VWAP (Volume-Weighted Average Price)

• Formula: Sum(Price × Volume) / Total Volume (intraday only)
• Above VWAP = bullish intraday
• Below VWAP = bearish intraday
• Institutional benchmark — they buy below VWAP, sell above
• Resets daily — intraday tool only

4. SIGNAL COMBINATIONS

Strong Buy Signal (High Confidence)

All of these together:

• RSI < 30 (oversold) AND recovering
• MACD crossing above signal line
• Price bouncing off lower Bollinger Band
• Volume increasing on the bounce
• Price above 200-day MA (long-term uptrend intact)
• Fundamentals solid (P/E reasonable, FCF positive, revenue growing)

Moderate Buy Signal

• RSI between 40-50 in an uptrend (pullback buy)
• Price touching 50-day MA support
• MACD histogram turning positive
• Volume above average

Trend-Following Entry

• Golden Cross (50-day crosses above 200-day)
• Price breaks above resistance with high volume
• ADX > 25 confirming trend strength
• Buy on first pullback after breakout

Strong Sell Signal

• RSI > 70 in a ranging market
• Price hitting upper Bollinger Band with declining volume
• MACD crossing below signal line
• Price reaching prior resistance level

RULE: Never rely on a single indicator. Always combine 2-3 minimum.

5. POSITION SIZING

The 1% Rule

Never risk more than 1% of total account on a single trade.

• Account: $1,000 → max risk per trade: $10
• If stop-loss is $2 below entry → buy 5 shares max
• Keeps you alive through losing streaks

Kelly Criterion

• f = (bp - q) / b
• b = win/loss ratio, p = win probability, q = loss probability
• Gives optimal position size based on historical win rate
• Use half-Kelly for safety (most professionals do)

Example

• Win rate: 55%, average win: $200, average loss: $100
• Kelly: f = (2 × 0.55 - 0.45) / 2 = 0.325 (32.5% of account)
• Half-Kelly: 16.25% — still aggressive, many pros use quarter-Kelly

6. EXIT STRATEGIES

Stop-Loss Types

• Fixed percentage: Sell if price drops X% (typically 5-8%)
• ATR-based: Stop at entry minus 2× Average True Range (adjusts for volatility)
• Support-based: Stop below key support level or moving average
• Trailing stop: Moves up with price, locks in profits (e.g., trail by 5%)

Take-Profit Methods

• Fixed target: Pre-set price target (e.g., 2:1 or 3:1 risk/reward)
• Technical target: Previous resistance level, Fibonacci extension
• Trailing: Let winners run with trailing stop

The 3 Hardest Sells

1. Cutting losses — "it'll come back" kills accounts. Honor your stops.
2. Taking profits too early — use trailing stops to let winners run
3. Holding through earnings — volatility spikes. Reduce position or hedge.

7. STRATEGY ARCHETYPES

Mean Reversion

• Theory: Prices revert to their historical average over time
• Entry: Buy when RSI < 30 or price 2+ standard deviations below mean
• Exit: Sell when RSI > 70 or price returns to mean
• Tools: Z-scores, Bollinger Bands, RSI
• Best in: Range-bound/sideways markets
• Weakness: Fails badly in strong trending markets (price keeps going)
• Key stat: Z-score above 1.5 or below -1.5 signals opportunity

Momentum

• Theory: Stocks that are going up tend to keep going up (and vice versa)
• Entry: Buy stocks with strongest 3-6 month performance
• Exit: Sell when momentum weakens (MACD crossover, RSI divergence)
• Philosophy: "Buy high, sell higher" — opposite of value investing
• Best in: Trending markets with clear direction
• Weakness: Sudden reversals can be devastating
• Key tools: Trend lines, MACD, RSI, relative strength vs index

Moving Average Crossover

• Entry: Buy on Golden Cross (50-day crosses above 200-day)
• Exit: Sell on Death Cross (50-day crosses below 200-day)
• Pros: Simple, catches major trends
• Cons: Very lagging — you'll miss the first 10-20% of a move and give back 10-20% at the end
• Variant: Use shorter MAs (9/21) for faster signals but more noise

Bollinger Band Squeeze

• Entry: When bands contract to minimum width, enter on the breakout direction with volume confirmation
• Exit: When price reaches opposite band or bands start contracting again
• Key: The squeeze only tells you a big move is coming — NOT the direction
• Must combine with: Volume, MACD, or other directional indicator

Pairs Trading

• Origin: Morgan Stanley, mid-1980s
• How: Find two highly correlated stocks (0.80+ correlation). When they diverge, go long the underperformer, short the outperformer
• Profit: When they converge back to historical correlation
• Market-neutral: Hedged against broad market moves
• Risk: Correlation can break permanently (one company's fundamentals change)
• Example: Coca-Cola vs Pepsi, Visa vs Mastercard

VWAP Reversion (Intraday)

• Entry: Buy below VWAP when overall trend is bullish
• Exit: Sell above VWAP or at end of day
• Best for: Day trading liquid stocks
• Why it works: Institutional traders use VWAP as a benchmark

8. BACKTESTING

The 4 Deadly Biases

1. Optimization Bias (Curve Fitting): Over-tuning to historical data. Fix: minimal parameters, out-of-sample testing
2. Look-Ahead Bias: Using future data accidentally. Fix: strict chronological processing
3. Survivorship Bias: Only testing stocks that exist today. Fix: survivorship-free datasets or recent data
4. Psychological Tolerance Bias: A 25% drawdown looks fine on a chart but feels devastating in real-time

Key Metrics

Expected Return per Trade: (Win% × Avg Win) + (Loss% × Avg Loss) — must be positive

Drawdown Math (Critical)

• 10% loss → need 11% gain to recover
• 20% loss → need 25% gain to recover
• 30% loss → need 43% gain to recover
• 50% loss → need 100% gain to recover
• The math gets exponentially worse. Avoiding big losses > chasing big wins.

Process

1. Define strategy rules with zero ambiguity
2. Get clean OHLCV data (Open, High, Low, Close, Volume)
3. Build engine (Python: backtrader, vectorbt, quantstats)
4. Split: 70% training / 30% out-of-sample validation
5. Run Monte Carlo simulation (randomize trade order) for robustness
6. Paper trade winners for 3+ months
7. Go live small — start with minimum position sizes

Michael Automates Backtesting Workflow (Proven)

His backtesting engine ($99) does this — we can replicate for free:

1. Get historical OHLCV data (from CCXT/Binance/Alpaca — free)
2. AI creates Python version of your trading strategy
3. Run backtest locally → get metrics
4. Compare with TradingView numbers to verify accuracy
5. If numbers don't match: Export TradingView Excel report, do trade-by-trade comparison
6. AI can auto-fix discrepancies by modifying the backtest engine code
7. Critical settings: Commission = 0.1%, timezone = UTC, date range = not full history

Auto-download data:

• CCXT pulls from Binance, Coinbase, Kraken automatically
• Store in /data/cache/ folder by asset and timeframe
• Format: BTC_USDT_1d.csv, ETH_USDT_4h.csv, etc.

Data Sources

• Free: Yahoo Finance (yfinance), Alpha Vantage, Alpaca, Polygon.io free tier, CCXT (100+ crypto exchanges)
• Paid: Norgate, Nasdaq Data Link (Quandl)
• Warning: Yahoo Finance has survivorship bias
• Best for crypto: CCXT + Binance (most history, most pairs)
• Best for stocks: Alpaca (free, clean data, built-in paper trading)

9. RISK MANAGEMENT

Core Rules

• 1% Rule: Never risk more than 1% of account per trade
• Daily Loss Limit: Stop trading if down 3% in a day
• Correlation Risk: Don't hold 5 tech stocks — one sector crash kills you
• Position Limits: Max 5-10 open positions at once
• Cash Reserve: Always keep 20-30% cash for opportunities

Circuit Breakers

• Hit daily loss limit → done for the day
• Hit weekly loss limit (5%) → reduce position sizes 50%
• Hit monthly loss limit (10%) → pause, review all strategies
• 3 consecutive losses → take a break, re-evaluate

Fees & Slippage (Often Ignored)

• Commission-free doesn't mean cost-free — there's still the spread
• Slippage: the difference between expected and actual execution price
• High-frequency strategies amplify these costs
• Always include fees + slippage in backtests or results are meaningless

10. COMMON MISTAKES

1. Trading without a plan → Write rules BEFORE trading
2. Ignoring position sizing → The 1% rule exists for a reason
3. Moving stop-losses → Set them and honor them
4. Averaging down → Adding to losers hoping they'll recover
5. Overtrading → More trades ≠ more profit (costs eat you alive)
6. Single indicator reliance → Always combine 2-3 indicators
7. Not accounting for fees → Especially in backtests
8. Skipping paper trading → Going live without 3+ months of validation
9. Revenge trading → Trying to win back losses with bigger bets
10. Ignoring the macro → Individual stocks don't exist in a vacuum

11. TOOLS & APIS

Broker APIs

• Alpaca — Commission-free, great API, paper trading built in (RECOMMENDED for us)
• Interactive Brokers — Most comprehensive, supports everything
• TD Ameritrade/Schwab — thinkorswim API

Data APIs

• Alpha Vantage — Free tier (5 calls/min), stocks + crypto + forex
• Polygon.io — Real-time and historical, free tier
• Yahoo Finance — Free via yfinance (unreliable, survivorship bias)
• Alpaca Market Data — Included with account

Python Stack

• pandas — Data manipulation
• numpy — Numerical computing
• ta-lib / pandas-ta — Technical indicators
• backtrader — Full backtesting framework
• vectorbt — Fast vectorized backtesting
• quantstats — Portfolio analytics
• scikit-learn — ML pattern recognition
• alpaca-trade-api — Broker integration

Architecture

[Data Feed] → [Indicator Engine] → [Signal Generator] → [Risk Manager] → [Order Executor]
     ↑                                                           |
     └──────────── [Backtesting Engine] ←────────────────────────┘
                          ↓
                   [Performance Analytics]

12. CRYPTO-SPECIFIC STRATEGIES (From Video Research)

Super Trend Strategy

• Built-in TradingView indicator
• AI can iterate: base version (44% P&L) → optimized V4 (3,605% P&L)
• Process: give AI base → "improve it" → backtest → repeat overnight
• Test on multiple assets to avoid overfitting

Strategy Tournament (Evolution Method)

• Create 10 different strategies with different parameters/logic
• Run all 10 in parallel with small positions ($100 each)
• After 1-2 weeks: cut losers, keep winners
• Create variations of winners, repeat process
• Natural selection for trading strategies

Market Regime Detection

• Run separate strategies for trending vs ranging markets
• AI detects current regime and switches strategies automatically
• Trend-following for trending markets, mean reversion for ranging

Prediction Market Strategies (Polymarket)

• 15-Minute Windows: BTC up/down resolution every 15 minutes
• Late Entry: Look at trend in last 3-4 minutes, enter in that direction
• Arbitrage: When both sides cost <$1 total = guaranteed profit
• Mention Markets: AI studies speech patterns to predict word usage
• Sports Scanner: AI researches obscure low-volume markets for edges
• Counter-Trend AI: Bet against AI consensus (contrarian)
• Wallet Analysis: Copy successful wallets → AI reverse-engineers their strategy
• KEY: Not dependent on bull/bear markets — works always

Multi-Agent Trading Architecture

• Coordinator: Delegates tasks, manages priorities, daily briefing
• Quant Scanner: Scans 30+ coins every 15 min with CCXT, confluence scoring
• Researcher: Daily news/intel scanning, deep dives
• Alert Agent: Formats and delivers signals
• Security Agent: 24hr audit for malicious code, prompt injections, bugs
• Each agent needs a detailed "soul" (personality + instructions)
• Agents work in parallel, not sequentially

Crypto Exchange APIs

• Hyperliquid — Decentralized perps, no KYC, good API
• Blofin — Centralized, API with trading perms (disable withdrawals!)
• CCXT Library — Open-source, supports 100+ exchanges, start read-only
• Alpaca — Stocks (commission-free, paper trading built in)

AI Strategy Iteration Workflow (KEY — From Michael Automates)

This is the single most powerful workflow from all our research:

1. Pick a base strategy (e.g., Super Trend, RSI+MACD, Bollinger Squeeze)
2. AI backtests it → gets baseline metrics (P&L, drawdown, win rate, Sharpe)
3. Tell AI "improve this" → it modifies parameters, adds filters, changes logic
4. AI creates V2, V3, V4... each iteration potentially better
5. Compare AI numbers with TradingView to verify accuracy
6. LET IT RUN OVERNIGHT — wake up to results
7. Score across multiple assets (BTC, ETH, SOL, SPY, AAPL) to avoid overfitting
8. Only keep strategies that work on 3+ different assets
9. Repeat until you have top 3 strategies → paper trade those

Michael's results: Super Trend went from 44% P&L → 3,605% P&L through this process.

Critical anti-overfitting rules:

• Test on multiple assets, not just one
• Use walk-forward validation (train on years 1-3, test on years 4-5)
• If a strategy only works on one asset, it's overfit — discard it
• Commission must be set to 0.1% (not 0%)
• TradingView timezone must be UTC for number matching

Strategy Tournament (From Alex Carter)

1. Create 10 different strategies (vary parameters, timeframes, indicators)
2. Run ALL 10 in parallel with small positions ($100 each)
3. Track for 1-2 weeks
4. Kill bottom 5 performers
5. Create mutations/variations of top 5
6. Run new tournament
7. Repeat — "evolution for trading strategies"
8. After 3-4 rounds, surviving strategies are battle-tested

Counter-Trend Against AI Crowd (From Coin Bureau)

• Most AI trading bots use similar data and strategies
• When AI crowd consensus is heavy one direction, bet the opposite
• Works because: AI herding creates overcrowded trades that reverse
• Requires: monitoring what popular AI strategies are doing
• Best for: short-term contrarian plays in crypto and prediction markets

Divergence Trading (From Coin Bureau)

• Use TBT Divergence indicator on 15-minute timeframe
• Contrarian entry when price makes new high/low but indicator doesn't confirm
• Specifically: price up but RSI/MACD diverging = weakening momentum
• Enter against the trend on confirmed divergence
• Best for: short-term crypto and Polymarket 15-min windows

Critical Crypto Trading Rules

• NEVER enable withdrawal permissions on API keys
• Whitelist your IP on centralized exchanges
• Store keys in .env files, never hardcoded
• Use testnet/paper trading first
• Rotate API keys every few months
• Run on dedicated machine (not personal computer)
• Daily security audit via cron job

13. POLYMARKET (PREDICTION MARKETS)

How It Works

• Binary markets: buy Yes/No shares in real-world outcomes
• Correct = $1.00 payout, wrong = $0.00
• Price = probability (Yes at $0.70 = market thinks 70% likely)
• Runs on Polygon (Ethereum L2), uses USDC.e
• Python SDK: py-clob-client
• Public data API (no auth): gamma-api.polymarket.com

Strategy Types

1. Arbitrage — Yes + No < $1.00 = free money (rare, need speed)
2. Late Entry — 15-min BTC windows, enter last 3-4 min when trend is clear
3. Mention Markets — AI analyzes speech patterns to predict word usage
4. Obscure Sports — Low-volume markets with inefficient pricing
5. Wallet Analysis — Reverse-engineer top performers' strategies
6. Market Making — Provide liquidity, profit from spread (advanced)

Why AI Has an Edge

• Can process more data than any human (speeches, news, stats)
• 24/7 monitoring of all markets simultaneously
• Niche markets have little competition = more mispricing
• Uncorrelated to stock/crypto markets — works in any cycle

Market Making (Advanced)

• Place both buy and sell orders, profit from the spread
• Requires: understanding inventory risk, dynamic spread adjustment
• Need algorithm to adjust quotes based on position, volatility, flow
• Complex math but AI handles research and implementation
• Start with wide spreads in low-volume markets, tighten as you learn
• Risk: getting stuck holding losing positions if market moves fast

TBO Cloud Strategy (Coin Bureau — Proprietary)

• Trending Breakout indicator on 4-hour timeframe
• Advanced moving average strategy — details not public
• Combined with TBT Divergence indicator for entry/exit
• We can build similar: EMA cloud (9/21/55 EMAs) with breakout confirmation
• Key concept: cloud acts as dynamic support/resistance, breakout above = long

Key Insight

Build scanner first (read-only, free). Paper trade. Then go live small.

14. CRYPTO vs STOCK STRATEGIES — WHAT TRANSFERS

Strategies That Work in BOTH Markets

Key Differences

What Needs Adjusting for Stocks

• Wider MA periods — crypto is faster, use longer MAs for stocks to filter noise
• Lower leverage — 2x max for stocks vs 5-10x crypto
• Market hours — need pre/after-hours data handling, overnight gaps
• PDT Rule — if under $25K, limited to 3 day trades per 5 business days
• Earnings events — stocks have scheduled volatility events quarterly
• Sector correlation — stocks in same sector move together more than crypto
• Volume patterns — stocks have predictable volume curves (open, close spikes)

Bottom Line

Yes, the core strategies transfer to stocks. The math is the same. What changes:

1. Parameters (slower timeframes for stocks)
2. Risk settings (lower leverage, wider stops)
3. Schedule (market hours vs 24/7)
4. Starting capital ($25K for day trading stocks due to PDT rule, or use swing trading to avoid it)

Best starting point for stocks: Alpaca (free, commission-free, paper trading built in, great API). Same strategies, just tune the parameters.

15. MULTI-AGENT TRADING DESK (From Coin Bureau)

Architecture

Each agent gets a detailed "soul" (personality + priorities + instructions). Without a soul, agents produce generic output.

Agent Roles

1. Coordinator (Betty) — Manager agent. Delegates tasks, manages priorities, monitors agent health, delivers daily briefing. Never does the work itself.

2. Quant Scanner — Scans 30+ coins every 15 minutes via cron job. Uses CCXT for OHLCV data. Runs indicators (RSI, MACD, Bollinger, etc). Produces confluence score for each signal. Rules: never places trades without operator approval.

3. Researcher — Runs daily. Scans news, X/Twitter, market updates. Produces morning research brief. Focused on narratives, sectors, new listings.

4. Alert Agent — Formats signals from quant scanner into actionable alerts. Pushes to Discord channels. Includes: asset, timeframe, direction, confidence score, entry/exit suggestions.

5. Security Agent (Radar) — 24hr security audit. Checks for malicious code, prompt injections, bugs. Quality assurance on all strategy code before deployment. Non-negotiable.

6. Backtesting Agent — Validates strategies against historical data before deployment.

7. Trading Agent — Executes trades on exchange APIs. Only activated after paper trading period.

Notification Pipeline (Discord Recommended)

• Private Discord server with bot admin privileges
• Bot creates channels per category (signals, research, alerts, P&L)
• Curated alerts — only coins/markets you care about
• Much better than Telegram (categories, threads, searchable)

Key Insight From Coin Bureau

"Your output is only as good as your input." Generic prompts = generic agents. Spend time writing detailed soul descriptions for each agent. Include:

• Exact responsibilities
• What data to use
• What format to output
• What rules to follow
• What to never do

Cost Reality

• Claude Max ($200/mo) — unlimited for heavy agent usage
• API credits alternative: $1,000-10,000/mo depending on volume
• Ollama (free) — can run specific tasks locally but less smart
• Strategy: use Claude for complex work, Ollama for routine scans

16. OUR APPROACH

Phase 1: Foundation

• Set up Alpaca paper trading (free)
• Build data pipeline for historical OHLCV data
• Implement core indicators: RSI, MACD, Bollinger, MAs, Volume
• Build backtesting engine with walk-forward validation

Phase 2: Strategy Development

• Backtest all 6 strategy archetypes against 5-10 years of data
• Find highest Sharpe ratio + lowest max drawdown combo
• Monte Carlo simulations for robustness
• Paper trade top 2-3 strategies simultaneously

Phase 3: Live (After 3+ Months Paper)

• Minimum position sizes
• AI-driven condition analysis (use Claude to interpret market context)
• Full risk management: 1% rule, daily limits, circuit breakers
• Dashboard monitoring all positions, signals, performance

Key Principles

1. Paper trade until proven over 3+ months
2. Position sizing > entry signals
3. Risk management is the ONLY thing that keeps you in the game
4. No single indicator is reliable alone
5. Past performance ≠ future results
6. Start with stocks, not crypto (more data, less manipulation)
7. Account for fees, slippage, and taxes ALWAYS