How the Jarvis supercomputer model works
Jarvis is Nhipbong's own quantitative model. Instead of offering gut-feel opinions, Jarvis works in numbers: the probability of a win, draw or loss, the most likely scorelines and the expected goals for each match.
What the model computes
For every match, Jarvis estimates the attacking and defensive strength of both teams, adds home advantage and the context of the competition, and derives the expected goals for each side. From that goal distribution, the model computes the probability of each outcome, the most likely scorelines, the chance of both teams scoring and the total goals in the match. This is a statistical approach widely tested in football analysis.
Why there is a "consensus" element
The final figure is cross-checked against independent reference data to raise confidence. When the sources point the same way, confidence is higher; when they diverge, Jarvis lowers its confidence and says so. We only publish the aggregated consensus figure, not any single source.
Transparency: predict first, compare later
Every prediction is recorded before kickoff, then compared openly against the real result on the Results page. This lets us measure the model with data across many matches, rather than only remembering the times it got it right. That is the difference between a model you can verify and a gut opinion.
Accuracy, measured in numbers
Calling a model "accurate" is easy; putting a checkable number on it is not. Two things need to be kept apart.
Pre-match predictions are the real test of the model, and every one is compared openly on the Results page, as the section above explains. Nothing is edited after the outcome is known.
In-play tracking is a different thing: Jarvis's live probability updates throughout the match, and in the closing minutes, once the game is largely settled, the side Jarvis favoured was correct in 814/853 recent matches, about 95.4%. To avoid any misreading: this figure only measures how closely the live number tracks the endgame, not the ability to call it in advance, because the nearer full-time the clearer the match, so the rate is naturally high. Step back to around minute 85 and it sits near 90.8%, reaching about 93.5% by minute 90, before rising as the result becomes certain.
In-play tracking figures (closing minutes)
What Jarvis does not do
A probability is not a certainty: a team with an 80% chance of winning can still lose, and that is part of the forecast. The numbers here are statistical analysis, not betting advice. Jarvis is Nhipbong's own model, not Opta, and has nothing to do with J.A.R.V.I.S from the Iron Man films.