Kristian Fagerlie asks GPT-6 Astra and Fable 5.1 to design strategies for short Bitcoin prediction markets, then compares their simulated performance. One strategy uses a simple directional entry with larger positions; the other considers distance from the target, volatility and remaining time before deciding whether to trade.
Kristian Fagerlie uses the recorded trial to illustrate why a higher win rate need not produce a better financial outcome. Position sizes, large drawdowns and accumulated fees materially affect the comparison. The results are specific to a short simulation and the prompts used, not evidence that either model provides a reliably profitable strategy.
Kristian Fagerlie then describes a small live trial and subsequent adjustment, acknowledging that favorable Bitcoin market movement explains part of the gains. The demonstration concerns the mechanics and limitations of automated experimentationWorkflow automation uses software or AI to complete a connected sequence of routine steps with less manual effort., not a recommendation to trade or a promise of steady returns.
Watch on YouTube
