Better machine learning and algorithms
We run machine-learning experiments most weeks, looking for gaps in the market. That work continues, alongside refinement of the algorithmic side: Poisson goal models, our own Elo ratings, consensus across models, historic trends and odds ranges. Each idea is measured against what already works and kept only if it adds something. The details are on the methodology page.
The latest AI models
AI is moving quickly. We plan to keep evaluating the newest models and using them to refine our choices and look for edges, as well as to keep improving the quality and readability of previews. New models earn their place in the pipeline by passing the same fact-checking and testing as the old ones, as described in how we build AI previews.