A model you can understand.

Daily ticket selection

The daily page compiles global fixtures by date, automatically queues each league’s season history, and evaluates available markets. Estimated confidence means the model probability of that market, not measured accuracy. It uses the last three games per team (all three required), at least at least 0.5 weighted venue-specific matches, and 20 recent league results for a baseline. Stability is checked by comparing the last two games with the last three. Recent-form and head-to-head requests must succeed, and other potentially material headlines exclude a pick for review. Injuries and coaching changes are not fetched or included in the estimates. It selects at most one fixture per league and never repeats a team. Combined estimates multiply probabilities assuming independence and must remain at least 60%. Ranking maximizes estimated likelihood greedily, not bookmaker value or proven profit. Some eligible games may be omitted because of these ticket-level constraints. Historical holdout calibration for these automatically selected markets has not yet been established.

From past results to match probabilities

MatchLens fits a simple independent Poisson goals model to completed results in the selected competition and season. Only completed matches dated before the prediction cutoff are included. The daily page also excludes today’s results when assessing future fixtures. Daily fixtures display in the configured timezone, default Africa/Johannesburg; the individual analysis page uses UTC.

Recent form, with a home advantage

The main form input is each team’s last three completed games, fetched across competitions for daily picks. A league baseline uses at most 60 matches from the last 180 days. Results within these windows receive exponentially decreasing weight with a 180-day half-life. The last three head-to-head meetings within three years have at most 6% influence, capped at a 0.25-goal change per team. Cross-competition opponent strength is not adjusted. Home scoring and away scoring averages form separate league baselines. A team’s home attack and defence are compared with its opponent’s away attack and defence. Five weighted prior matches at league average smooth small samples.

Three matches are a small sample: estimates can change sharply after one result and are not measured success rates.

What the numbers mean

The resulting expected goals generate score probabilities. Summing those scorelines gives home win, draw, away win, over 2.5 goals, and both teams to score. These expected goals come from historical scores; they are not shot-based xG. Outputs describe 90 minutes plus stoppage time, excluding extra time and penalty shootouts.

Data quality comes first

At least ten earlier competition results are required, and each team must have an earlier match. Effective venue-specific sample size determines the Limited, Moderate or Broad data-depth label. Limited is below three weighted matches; Moderate is below five. This label is not a claim about accuracy. Newly promoted teams, cups with little history, neutral venues and stale data can produce weak estimates.

What this version does not know

Team-name-matched BBC Sport headlines about postponements, weather, strikes, boycotts or outbreaks screen automatic tickets for review. Injury and coaching records are not fetched or used. These checks do not numerically adjust the model. News coverage is incomplete, and no matched report does not prove that a team is unaffected. Confirmed lineups, tactical changes, transfer impact and bookmaker odds are not numerical inputs. Third-party predictions are never fetched. Goals are treated as independent, so low-scoring draws may be misestimated. Expected goals are constrained to 0.15–5.0 per team. Scorelines from 0 to 25 goals per side are evaluated and outcome probabilities are normalized.

Evaluate before relying on predictions

No accuracy percentage is claimed. The included command-line backtest compares rolling predictions with a past-only league-average baseline using multiclass Brier score and log loss. A retrospective test can still be optimistic because historical provider data may have been corrected after the match. Better scores on one season do not guarantee future performance.

Online data

API-SPORTS supplies raw fixtures worldwide, recent results and head-to-head meetings. The daily updater discovers competitions automatically. Data is refreshed by the administrator or a cPanel scheduled job, not on every visitor request. Coverage and accessible seasons vary with the provider and subscription. The app cannot cover competitions the provider does not supply.