Inside the Trade · 10-part field guide
Understand the market before you automate it.
Follow one crypto order from the book to the fill, then learn how researchers test an idea and how production systems keep state when markets move.
Part 01 · Start here
What Happens After You Tap Buy? Anatomy of a Crypto Trade
Follow the orderRead the market
Orders, liquidity and the roles that create a price.
Test the idea
Data, model targets and validation without hindsight.
Survive production
Systems, venue differences and the cost of delay.
Applied layer · AI agents
Now give the system a model—and less authority than it asks for.
Build on the market foundation with structured proposals, deterministic risk gates, paper execution and an audit trail.
Start the AI-agent guide
Understand the agent
Automation changes who decides. It does not remove responsibility.
Learn what an agent is, how it differs from older automation, and why permission limits matter more than a clever prompt.
AI Trading Agent vs Trading Bot vs Copy Trading: What Actually Changes?
AI agents, trading bots, and copy trading automate different decisions. Compare who creates the signal, who controls execution, and where the risks sit.
Read guide
The Trade an AI Agent Must Refuse: Seven Safety Gates Before Execution
A safe trading agent needs explicit reasons not to trade. These seven external gates turn vague caution into testable rejection rules.
Read guide
Who Signed the Order? Wallet Permissions for AI Trading Agents
A model should never hold unlimited signing authority. Learn how keys, scoped permissions, policy checks, approval, and revocation fit together.
Read guidePrediction-market lab
Build the experiment before you believe the result.
Polymarket gives researchers public prices, books, trades, and market rules. This series turns them into a paper-trading lab—without pretending a forecast equals an executable edge.
Lab note 01
How to Build a Polymarket AI Trading Agent—Paper Trading First
Build a useful Polymarket AI agent without giving it a wallet: collect public data, make time-stamped forecasts, simulate realistic fills, and audit every decision.
Lab note 02
Can AI Beat Polymarket? A Time-Locked Trading Experiment
A reproducible experiment for testing AI forecasts against Polymarket odds without leaking the answer, cherry-picking markets, or pretending paper fills were real.
Lab note 03
When the Winning Bet Still Loses: Spread, Slippage and Partial Fills
A correct prediction is not automatically a profitable trade. Learn how bid-ask spread, visible depth, fees, latency, queue uncertainty, and partial fills change the result.
Lab note 04
Crypto Charts vs Polymarket Odds: Two Markets, One Event
Crypto price charts and prediction-market odds can react to the same event while answering different questions. Here is how to compare them without forcing a false signal.Test the method
A score is only useful when the test can survive scrutiny.
Freeze the data, count execution costs, defend the tool boundary, and publish failures alongside apparent wins.
How to Backtest an AI Crypto Strategy Without Fooling Yourself
A practical framework for testing AI crypto strategies with point-in-time data, frozen models, realistic costs, walk-forward validation and a hard boundary before live capital.
Prompt Injection for Trading Agents: Can a Headline Hijack a Bot?
A defensive guide to keeping hostile headlines, webpages and tool output from turning an AI market analysis task into an unauthorized trade.
When AI Agents Trade Against Each Other: Who Makes the Market?
A practical explanation of how automated market makers quote prediction markets, manage inventory and survive adverse selection—without pretending the order book reveals who runs an AI agent.
AI Trading Agent Scorecard: How Boxmining Will Test Models
Boxmining’s methodology template for evaluating AI trading agents across forecast calibration, drawdown, execution, refusals, safety rules and reproducibility—not headline profit alone.
Chart-reading foundations
The old tools still matter.
An agent does not make support, resistance, structure, or invalidation disappear. These original Boxmining guides teach the chart language behind many automated signals.