The experiment starts with historical data for Kalshi's 15-minute Bitcoin up-or-down market. A structured research skill asks Codex to inspect unusual patterns and produce five hypotheses without testing them prematurely, while a separate evaluation skill handles backtestingBacktesting applies a strategy to historical data to estimate how it would have behaved under past market conditions., walk-forward checksWalk-forward validation repeatedly trains or selects a strategy on earlier data and evaluates it on the next unseen time period. and out-of-sample analysisOut-of-sample testing evaluates a model or strategy on data that was not used to create, select or tune it..
Three ideas survive the initial screening, with a volatility-persistence rule selected for a live trial. The rule requires the previous contract to move at least 15.4 basis points, reads the current market midpoint after five minutes and chooses the favored side before attempting a small fill-or-kill order one minute later.
Codex writes the execution codeAn AI trading bot uses AI-assisted analysis or decision logic to generate and execute trades under defined rules and controls. around the documented strategy and connects it to Kalshi through WebSockets instead of polling. The creator verifies that the service is healthy, that timing and market-depth checks are active, and that windows are skipped when the previous Bitcoin move does not meet the volatility threshold.
After roughly a day, the account rises from about $120 to $137, with 14 winning and four losing trades in the reported run. The video treats this as a fast experiment rather than proof of a persistent edge, noting that the historical evidence did not support strong confidence in long-term returns.
The broader lesson is that reusable agent skillsAn AI agent skill is a reusable package of instructions, resources, and tool guidance for performing a bounded kind of work. can separate idea generation, evaluation and implementation while preserving an auditable strategy specification. Promotional references to the creator's fund concept and affiliate links are omitted.
Watch on YouTube
