How we
work.

For every scheduled game in our covered leagues, an AI model produces a probability split between the two sides. The model uses web-search-grounded analysis at the time of prediction, considering recent form, lineup news, rest, schedule context, head-to-head history, and sport-specific factors like starting pitcher in MLB or goalie matchup in NHL.

We record the probability and the reasoning at the moment of prediction. Nothing is rewritten after the fact.

A background monitor runs continuously, polling each scheduled game once it should have ended. When a game finishes, the result is fetched from public scoring data and stored alongside the prediction.

Every prediction is binary: the model favored a side, the actual winner either was that side or wasn't. The model is "right" or "wrong" per game — no half-credit, no spreads, no shifting goalposts.

The key insight: a probability isn't just a guess at the winner. It's a claim about confidence. A model that says "80%" should be right 80% of the time. Not 90%. Not 60%.

We bucket predictions by confidence band (50–60%, 60–70%, 70–80%, 80%+) and compute the hit rate in each. When the actual hit rate diverges from the expected midpoint, the model is miscalibrated — and we show you the delta. Future predictions in that band get auto-adjusted.

This is "learning" in the only sense that actually matters: the system gets sharper with data.

Fourthsight does not accept bets. We have no integration with sportsbooks. We do not earn affiliate revenue from operators. We make no claim that following our predictions will produce a profit or any other financial outcome.

Our goal is narrower and more honest: show how often a model is right, and let you decide what to do with it.

NBA · NHL · MLB · English Premier League · UEFA Champions League · major professional boxing. We'll add more leagues based on what the early users actually engage with.

Fourthsight is built by a small team. The forecasting layer is powered by Anthropic's Claude. The monitor runs continuously on a managed VPS. The frontend is what you're looking at.

Questions, feedback, ideas, complaints: hello@fourthsight.app.