The Performance Gap: Diagnosing Why Backtested Crypto Agents Consistently Underdeliver at Deployment
The distance between a backtest showing forty percent annualized returns and an agent that captures four percent in live markets is not a calibration error — it is the predictable outcome of statistical assumptions that cannot survive contact with real order books, real market participants, and real execution constraints. This analysis identifies the specific mechanisms through which backtesting environments systematically overstate agent performance, and offers a structured framework for buildi