Data & methodology

• Historical matches (2013–14 to 2025–26): a curated set of turning-point fixtures with final and half-time scores, compiled editorially from the public record. Line-ups, events and detailed stats are not in this dataset.
• Squads, managers and season outcomes: compiled editorially from public records, with approximate join and leave dates.
• Current season (2026–27): results and upcoming fixtures gathered from public web sources on 26 October 2026. Kick-off times may be missing. Please verify against official sources.
• Licensed provider data (fixtures, line-ups, events) is integrated behind a feature gate and stays switched off until the provider contract covers this use.

Player ratings (0–100) and attributes are BeTheReds estimates for a given point in time. They are not official, and they are not taken from any licensed rating system. Opponent strength, budgets and revenue figures are estimates too, rounded to an order of magnitude.

Matches are simulated possession by possession. Team strength comes from the selected players, their fit for each position, roles and tactical instructions. Each possession can lead to a chance, and each chance has an expected-goals value. Outcomes are then drawn at random using those probabilities, with fatigue, cards, injuries and substitutions applied as the match goes on.

Seasons are simulated fixture by fixture, and careers add transfer negotiation (which can fail), contracts, finances and player development.

The AI (a large language model) is never part of the engine. It receives the engine's output and writes explanations, commentary and advice.

The engine is checked against long-run Premier League patterns and avoids dominant strategies that win no matter what:

• about 2.75 goals per match
• about 28% of matches drawn
• a realistic home advantage and shot conversion
• season points distributions that match how strong each squad was

Every simulation stores the data snapshot ID, engine version, rules version, random seed and your decisions. The same inputs and seed always give the same result on the same engine version. Use "Verify reproducibility" on any result to re-run it.

Historical simulations only use what was known on the start date. A career that starts in 2013 does not know how any player turned out later. Future development is probabilistic, and once you change history, later seasons play out in your alternate timeline instead of following real events.

One simulated match is one roll of the dice, so we encourage 100 or more runs and show distributions and confidence ranges. Even then, a simulation is a hypothetical scenario. It does not show that a decision would have changed a real result, and it is not a betting forecast.