Results and opponent strength
An Elo rating updates after each result, taking the opponent's rating into account. Points per game, goal difference, goals scored and goals conceded add information about overall performance.
Matches
Fixtures, results and PitchNexus predictions in one place.
How our predictions workThe information behind the percentages, and how to read them.
PitchNexus compares each team's recorded match history to estimate three possible outcomes: a home win, a draw, or an away win. These are statistical estimates, not promises about what will happen.
What goes into a prediction?
Recent form
The model looks at results and goal difference from each team's latest five recorded matches, or the available matches when there are fewer than five.
Home and away record
It compares the home team's home results with the away team's away results and includes a modest baseline home advantage. The home designation is taken from the schedule.
Shared opponents
When both teams have faced the same opponents, the model compares those performances. This adjustment gets less weight when there are few shared opponents.
These signals are combined into a difference in estimated team strength. A probability formula converts that difference into home and away chances, with a separate draw estimate that increases when the teams appear evenly matched. The three probabilities add up to 100%, apart from display rounding.
How to read the percentages
Illustrative example — not a live match
55%Home win
25%Draw
20%Away win
The home team is favored, but the model still assigns a combined 45% chance to a draw or an away win. The highlighted outcome is simply the one with the highest estimated probability.
Two different kinds of information on each card
How strongly the model favors an outcome
- CLEAR: the leading outcome has at least 70% probability and leads the next outcome by at least 30 percentage points.
- LEAN: it does not qualify as CLEAR, but the leader has at least 55% probability and a lead of at least 15 percentage points.
- CLOSE: neither threshold is met.
These labels describe the predicted gap. They do not tell you how much match history is available.
How much supporting evidence is available
- STRONG: both teams have at least five recorded games, at least two shared opponents, and relevant home/away history.
- MODERATE: not STRONG, but both have at least three games and either a shared opponent or relevant home/away history.
- LIMITED: neither higher level applies, but both have played at least once and have at least three games combined.
- MINIMAL: the remaining cases, including a team with no recorded games.
Relevant home/away history means at least one home game for the home team and one away game for the away team. These match-specific labels differ from the evidence labels on Team Analysis.
How the models are tested
The research process compares a baseline model with versions that add home/away results, shared opponents, or both. The current production model combines both enhancements.
- Replay historical matches in time order. Build the prediction from earlier match history, then add the result after making the prediction.
- Compare four successive test windows. This walk-forward approach checks performance across different periods rather than relying on a single split.
- Check both picks and probabilities. Accuracy measures how often the highest-probability outcome occurred. Brier score measures how far the probabilities were from the actual outcome; log loss particularly penalizes confident mistakes. Lower is better for both probability measures.
Historical validation does not use final standings or the finished Team Analysis scores to predict earlier games. Research results are reviewed; rerunning validation does not automatically replace the production model.
What the model cannot know
It does not currently use player availability, injuries, starting lineups, training progress or weather. It only knows the match history recorded in the system; missing results and limited history can affect the estimate. Home designation may not reflect a meaningful advantage at a neutral venue.
Small samples reduce the weight of performance adjustments. Teams without recorded results start with neutral rating defaults, so a prediction can still appear with minimal evidence. Evidence strength is not a guarantee of accuracy, and a strong favorite can lose.
Rankings and predictions answer different questions. Rankings compare overall performance within a group. Predictions estimate the outcome of a specific matchup, so a ranking position alone does not determine the favorite.
Predictions are regenerated when PitchNexus completes a data refresh; they are not live in-game forecasts. This explanation describes the production model reviewed on September 29, 2026.