Automatic analysis
Choose a fixture to collect evidence.
Select a fixture above for automatic analysis. Results, published odds and attributed expert forecasts are fetched automatically where available. Advanced inputs and fictional demo are optional.
Advanced inputs and demo
Use a podcast transcript
One collective Totally Football Show opinion. Paste an available transcript; speaker names are optional. Only pre-match evidence counts.
How it works
The main prediction chooses the outcome with the most independent human votes, then the most frequently published score within that outcome. Tied outcomes use odds, then statistics; tied scorelines use statistics. Agreement is a vote share, not a calibrated probability. No human forecasts means no opinion-led pick. The model below is a separate comparison.
The selected statistical model fits opponent-adjusted attack and defence, home advantage, decayed goals/xG and a Dixon–Coles low-score correction. Its parameters and probability calibration are chosen using chronological validation. Small demo inputs use the original venue model. Outcome probabilities blend statistics (65%), an average of margin-adjusted bookmaker prices (25%) and deduplicated human expert consensus (10%). Published goal markets can partially calibrate goal rates; this layer is experimental. Source-blend weights and confirmed-lineup/rest effects are learned only after enough real pre-match snapshots pass chronological holdout checks. Reporting headlines are context unless a dated, attributed match forecast can be extracted. Missing inputs redistribute those weights. Expert exact scores also contribute a provisional 25% mixture of smoothed score distributions. Verified pre-match community polls can adjust outcomes by up to 5%. When verified Opta probabilities are available, base weights are statistics 55%, odds 25%, experts 10%, Opta 10%, before score-level opinion. Opta also supports tied human outcome choices. Weights are experimental.
Use timezone-aware dates. Only results, odds and opinions recorded before kickoff qualify. Opinions need source, url, published_at and outcome (home/draw/away); use the same origin identifier for repeated reports. Only explicit journalist predictions count numerically; general headlines are never treated as votes.