A good prediction doesn't always make a good decision

I started developing an FPL tool because I kept returning to a single question: what is the purpose of predicting the best player if signing them does not ultimately result in the best decision? The complexity soon increased. You have a budget, transfer limits, players you currently possess, upcoming fixtures, and the occasional disastrous decision made several gameweeks prior.

To address this, I began creating the FPL Decision Engine. The objective was to merge football data, AI reasoning, and rule-based checks to evaluate real-world decisions instead of merely ranking players. One aspect that particularly captivated me was distinguishing between what the AI suggests and what the system permits. An AI model may recommend an outstanding player, but if the transfer surpasses the budget or violates the three-players-per-club rule, the suggestion must be rejected.

Confidence alone does not validate an invalid decision. I also began investigating another aspect: what transpires when my judgment diverges from the AI's? Does it genuinely enhance my decisions, or simply heighten my confidence in them? This research is still in progress, and I have not yet determined if the engine improves FPL performance.

There is still much testing to be done. The publicly accessible repository includes a refined version of the code, architecture, and a practical example. While the project began as a football endeavor, it has piqued my interest in decision systems in general. Predictions are valuable. However, determining what to do with them proves to be the more challenging aspect.