★ Premium reports · high-conviction hit rate · 471,853 all-time
69.4%
+36pp vs randomPremium subscribers get only the model's highest-conviction reports — fixtures where the top probability clears 55%. Random guessing scores 33%; always-home scores ~43%. Premium reports hit 69.4% across 471,853 graded fixtures. Free tier shows every fixture at 53.1% baseline accuracy.
All-fixture backtest baseline
Backtest
53.1%
1,520,107 · +20pp
Brier
0.580
↓ 0.667 unif.
Log-loss
25.689
↓ 1.099 unif.
Live since v4.2: 56 graded · 5.4% · Brier 0.018.
★ Premium · Live stage updates
Free tier locks in the pre-match analytics report. Premium re-runs the model at three decisive moments — and surfaces only the reports that meaningfully shift.
~1h before kickoff
Re-runs with the confirmed XI — ages, injuries, key absentees factored into the final probabilities.
At HT whistle
In-play update from the live scoreline + shot counts. Often flips the projected outcome when the data disagrees with the goals.
Late-game window
Final swing call — chase, hold, or upset alert when momentum shifts in the final 15.
Honestly labelled by what each era contributes
Ratings foundation
National-team strength priors · ungraded, context only
Modern training
Walk-forward · per-league calibrated
Live graded
SHA256-locked · v4.2
Faroese top flight to Brazilian fourth tier — every era honestly labelled by its role in the pipeline.
See Premium Plans →Active championships
Top 6 by accuracy · min sample 259+
Vertical line = 33% random baseline.
See Premium Plans →
SHA256-locked before kickoff·graded automatically·public & verifiable
See live analytics reports