Pair AI-assisted research with our Make It Simple But Significant trading framework — multi-timeframe analysis, zone identification, and proven setups.

Cross-reference AI insights with live charts on TradingView. Follow the Dreams LLC profile for curated analysis and market perspectives.

Install Claude Code from claude.com/claude-code. From here on, you talk — it types. Every step below becomes a sentence you say to it, starting with: "Help me install Bullpen's CLI so we can build a trading bot."

Sign up at Bullpen, then have your AI install its command-line tool and log in. This creates the wallet your bot will trade from — the keys stay encrypted on your machine, never in the code.

Decide before any code exists: who to copy (top traders by profit, never raw volume), how much per trade, and your daily maximum. Small fixed sizes ($5) and a hard daily cap ($50) are the novice-safe defaults we use.

Tell your AI to build in: a price band (skip anything under 5¢ or over 95¢), a slippage limit (never pay more than ~2% above the copied price), per-trader spending caps, and skipping live in-game markets. These rules are the difference between a bot and a leak.

The killer feature: before copying anyone, replay their recent trade history and check it would have made money at your size. Big-profit traders often don't copy well. Ours rejected the #4 trader on the leaderboard — his history would have lost 71¢ per dollar.

Have your AI build a local dashboard with one start/stop button and a live log of every decision. Run it with zero balance first — watch what it would do for a few days before a dollar is at risk.

Deposit an amount you can lose without pain — $50–$150 to start. Press the button. Review the log daily for the first week, and let performance data, not excitement, decide when to scale.

Install Claude Code from claude.com/claude-code. Your first sentence: "Build me a paper-only bot that watches Polymarket's 5-minute BTC Up/Down markets for a combined-ask edge under a dollar."

"Buy equal shares of Up and Down when their combined ask price, plus estimated fees, leaves at least a few cents of profit per dollar pair." One sentence, no ambiguity, before any code exists.

Tell your AI builder the bot must never load a private key or call a signed order-placement endpoint during this phase — public market-data reads only. This is what makes the whole build genuinely zero-risk from day one.

A minimum net-edge threshold (skip anything too thin to survive fees), a per-pair position cap, a hard daily paper-spend cap, and a cutoff near each market's close — no opening a position with no time left for it to matter.

Run it against live order-book data with pretend capital and find out, honestly: does this edge actually appear often enough to matter? Does it survive Polymarket's real fees? Not assumed — measured.

One start/stop button, open positions, closed trades, realized P&L. Every skipped opportunity logged with its reason — a thin edge you didn't take is exactly as informative as one you did.

Going live — loading a real key, placing real orders — is a separate, deliberate step, made only after paper results earn it. Plenty of edges that look real on paper don't survive real depth and real fees. Finding that out costs nothing here.

Install Claude Code and Python. Then every step below becomes a sentence you say, starting with: "Build me a paper-trading bot that goes long BTC when its 10-day EMA crosses above its 21-day EMA on the daily candle."

An EMA is just an average of recent prices that reacts faster to new days. Fast average above slow average = the trend is up, be in. Below = trend is down, be in cash. If you can read two lines on a chart, you understand this bot completely.

Have your AI download years of daily prices and replay the rule: how would it have done vs just holding? Through the last crash? Demand honest answers including fees. If the numbers don't convince you, stop here — that's the system working.

Start the bot with $100,000 of pretend capital. It fetches real prices and makes real decisions — only the money is simulated. This is the single most important safety rule: the bot earns trust with fake dollars before it ever touches real ones.

One local web page with a START/STOP button, the current signal for each coin, an equity chart, and a log of every decision with its reason. If you can't explain why the bot did something, it shouldn't be doing it.

Have your AI register the bot with Task Scheduler so it launches at every logon and checks the market every hour. From now on it runs while you sleep, work, and live — that's the entire point.

Let it paper-trade for weeks. Compare its equity curve against just holding. Going live afterwards is a separate, deliberate decision — a real exchange account, trade-only API keys, and small money you can afford to lose. Many people rationally never take that step.

Install Claude Code and Python. Your first sentence: "Build me a paper-trading bot that watches my stock watchlist and alerts me when a 10/21-day EMA cross happens, with the reasoning."

Use stocks you actually know and follow — your existing broker watchlist is perfect. A bot watching companies you understand teaches you something; a bot watching random tickers is noise.

"Long while the 10-day EMA is above the 21-day EMA on the daily close; cash otherwise." Written down before any code exists. If a rule can't be stated in one sentence, a bot shouldn't trade it.

Here's what ours found, and you should expect the same: on big steady stocks, momentum often loses to simply buying and holding — it wins by cutting crash damage, not by beating bull markets. Know this before you start, not after.

The bot manages pretend capital and notifies you at every cross with the why. You get the full experience — signals, discipline, drawdowns — at zero risk, while your own judgment sharpens with every alert.

Add the strategy script to TradingView and flip through your stocks. Seeing two years of crosses on a chart teaches you more about whipsaw and trends than any article ever will.

After weeks of paper results: maybe the bot becomes your signal service and you place occasional trades yourself; maybe it stays a learning tool. Both are wins. Automated real-money orders are a far bigger step — and never a requirement.

Install Claude Code from claude.com/claude-code. Your first sentence: "Build me a paper-trading bot that fuses a national forecast, a model ensemble, and live airport readings into a daily high-temperature model, and trades the gap against Kalshi's weather markets."

The national forecast gives a single estimate, the ensemble gives dozens of scenarios, live station readings give what's already been observed today. Average the first two, then ratchet up to the observed max — today's high can't resolve below a reading already recorded.

Counting ensemble members directly produces false 0%/100% probabilities the moment a bucket falls outside the observed scenarios — and a false 100% is exactly the input that produces a max-size bet on a loser. Dress each member with a shrinking statistical kernel instead.

A minimum model-confidence floor (skip anything the model itself calls a longshot), a price band, a per-city exposure cap (every bucket in one city's market settles off the SAME reading — treat them as one bet, not several), and fee-aware position sizing tuned off real win rate.

Before booking a paper trade, walk the real order book and cap the position to what it can genuinely supply at that price. Cheap longshot buckets often support a fraction of the size a fixed-budget bet wants — skip the trade rather than pretend a fantasy fill is real money.

Log the model's stated probability alongside the outcome for every closed trade. A calibration score and a stated-vs-actual table tell you whether the model is honest about its own uncertainty — win rate alone can't, since it conflates model quality with the price level traded.

Paper-trade for weeks, then audit honestly: is realized profit spread across many trades, or sitting almost entirely in one longshot a real order book couldn't have filled at that size? Our own first audit found exactly that — the fix is what any live decision should wait on.

A bot that follows proven Polymarket traders and copies their moves with strict spending limits. The exact blueprint behind our Bot 01 — prediction markets.

A bot that buys BTC, ETH & SOL uptrends and sits safely in cash through downtrends. Free data, free tools, pretend money until you say otherwise.

The same momentum engine pointed at the stocks you already follow on your broker — with alerts that tell you when to buy and, more importantly, why.

A bot that hunts for moments when buying both sides of a market costs less than the guaranteed $1 payout — no direction to predict, just arithmetic. Reads public data only; never risks a cent while paper trading.

Fuses a national forecast, a 31-member model ensemble, and live airport readings into a daily high-temperature model, then trades the gap against Kalshi's own city-by-city markets. Reads public data only; never risks a cent while paper trading.