Methodology

Prognosist is built around one principle: predictions should be explainable. We do not publish "guaranteed picks" or editorial guesses. Every forecast is produced from structured football data, statistical modelling and market comparison.

1. Data sources

Prognosist uses football data from third-party providers, including API-Football where available. This may include fixtures, results, scores, lineups, events, team statistics, player information and league tables.

We store key data points in our own database to improve consistency, reproducibility and auditability. This helps us understand which information was available when a prediction was produced.

Odds data may be collected from a panel of major bookmakers through third-party feeds. We use these prices to estimate market-implied probabilities and compare them with model probabilities.

Bookmaker odds are used for analytical comparison only. Prognosist does not accept bets and does not operate as a sportsbook.

2. Team strength index

Each team receives a continuous strength index that is updated as new match data becomes available.

The index combines several signals, including:

• attacking output
• defensive performance
• goals scored and conceded
• expected goals for and against
• shots and shots on target
• recent form
• home and away performance

The strength index is league-relative. This means that a team's number is mainly meaningful inside its own league context. A Premier League team and a Bundesliga team should not be compared directly using only the raw index.

The purpose of the strength index is not to rank every club in world football, but to help estimate how strong a team looks relative to its regular competition.

3. Probability models

Prognosist maintains statistical probability models for football markets.

The baseline model estimates probabilities for outcomes such as full-time result, total goals and Both Teams to Score. It uses team strength, attacking and defensive indicators, recent form and historical performance.

We also test more advanced model variants that use expected goals and additional contextual features. These variants are validated through backtesting before being promoted to production.

Each model outputs probabilities, not certainties. A team with a 60% win probability is still expected not to win many times out of 100 similar matches.

Predictions shown on Prognosist are model-based estimates, not editorial guarantees.

4. Value-bet detection

For supported markets, Prognosist compares model probabilities with market-implied probabilities from bookmaker odds.

First, the bookmaker odds are converted into implied probabilities. Then the market margin, also known as overround, is adjusted where possible. If the model probability is higher than the adjusted market probability by a defined threshold, the platform may label that outcome as potential value.

A value label does not mean that the bet will win. It means that, according to the model, the price may be higher than the estimated probability suggests.

This is an analytical signal, not betting advice.

5. Backtesting

Before significant model changes are promoted, they are evaluated on historical data.

We monitor several metrics:

• Brier score — measures probability calibration. Lower is better.
• Log-loss — penalises overconfident wrong predictions.
• Closing-line behaviour — checks whether value-labelled selections often move in the expected direction before kickoff.
• Historical ROI simulation — estimates how a flat-stake strategy would have performed in the past.

These metrics help us understand whether a model is better calibrated than the previous version. However, backtest performance does not guarantee future results.

Football changes. Markets change. Model performance can decline over time.

6. Markets covered

Prognosist currently focuses on the most common football prediction markets:

• 1X2 full-time result
• Over/Under total goals
• Both Teams to Score

Additional markets may be supported with limited or developing coverage, including:

• Asian Handicap
• first-half markets
• second-half markets

Some markets, such as team totals, corners and cards, require deeper data and larger samples. We do not treat these markets as fully calibrated unless the data quality is strong enough.

7. Known limitations

No football model can capture everything.

Important limitations include:

Smaller leagues

Data quality and xG coverage may be weaker outside major competitions. Where the sample is limited, confidence should be treated as lower.

Lineups and injuries

Late injuries, rotation and tactical changes can significantly affect a match. Official lineups are usually available only close to kickoff, and pre-match predictions may not fully reflect last-minute changes.

Cup competitions and friendlies

Knockout matches, international cups and friendlies are often noisier than regular league fixtures. Motivation, rotation and tactical priorities can be harder to model.

Market movement

Odds can move quickly. A value signal may change if the market price changes after the prediction is published.

Live betting

Prognosist predictions are primarily pre-match. We do not currently maintain a live in-play prediction model.

AI-generated explanations

Some previews and explanations may be assisted by AI. We aim to ground this content in available data, but errors or omissions may still occur.

8. Updates and transparency

We aim to publish significant methodology changes, model-version notes and backtest summaries when major updates are released.

If you notice a possible calibration issue, data mistake or methodology question, please reach out through our contact form and pick the Bug report or Feedback or feature idea subject so it lands in the right inbox.

For important legal information, please also read our Disclaimer, Terms of Use and Responsible Gambling pages.