General
BestPreviews is a platform that provides detailed match insights including previews, live match data, lineups, commentary, and statistical analysis to help users understand football matches better.
Stand Out Games
No. Quality measures calibration (how accurate the probability spread was), not profit or betting ROI.
A model can have strong average quality while still losing on “pick the favourite” style bets, and vice versa. Use Our beats market to see how often our full distribution beat the market’s on finished games in this stand-out set.
After full time we know the actual outcome (Home win, Draw, or Away win). We take:
- Market implied probabilities (Home / Draw / Away %)
- Our adjusted probabilities from the preview model
For each outcome we measure how far the predicted probability was from reality (1 for the outcome that happened, 0 for the others). Those errors are combined into one Brier score per model. Lower Brier = better.
We then display:
- Our prediction — our quality %
- Market prediction — market quality %
The card text explains who was closer and by how much on the actual outcome’s probability.
Avg our quality is the average of our model’s prediction quality across finished matches on this page.
For each completed game we compare our Home / Draw / Away probabilities to the actual result using a standard Brier score (a proper scoring rule for probabilistic forecasts). That score is converted to a 0–100% quality rating, where higher means our probabilities were better calibrated to what happened.
The summary chip is simply the mean of those per-game “Our prediction” percentages. It answers: on average, how well did our full 1X2 distribution match reality? — not whether we picked the winning outcome.
Our beats market is the percentage of finished games where our model was closer to the actual result than the market, using the same Brier-based quality measure.
Example: if the chip shows 62%, our probabilities were better calibrated than the market’s in 62% of the finished stand-out games in the current view.
This is not the same as “we predicted the winner more often.” A model can assign a higher probability to the correct outcome (better Brier) even when its favourite pick (highest % outcome) was wrong.
Only games listed on Stand Out Games: football previews with market movement data and max |Δ| ≥ 0.5% (meaning our model meaningfully shifted at least one outcome vs the market).
When you change the phase filter (All vs Post Match), the set of games — and therefore the averages — updates to match that view.
Those two chips only appear when there are finished matches with scores in the current list, and when you’re on the All or Post Match filter.
Pre Match and Live tabs focus on games still in progress or not yet played, so there is no final result to score prediction quality against.
State Analysis
Form Analysis shows how a team has performed in recent matches, typically based on wins, losses, draws, and consistency.
Market Movement reflects changes in team performance trends or expectations over time, often influenced by recent results, injuries, or form shifts.
Possession Efficiency measures how effectively a team uses ball possession to create chances, rather than just holding the ball.
Team Power Ranking is a performance score (0–100) that reflects a team’s overall strength based on recent form, stats, and match performance. Higher scores indicate stronger teams.