StatsBrain.ai
Public accuracy report

How accurate are our match projections?

StatsBrain publishes pre-match simulations for football fixtures worldwide. This page grades those projections against real final results — openly, with sample sizes disclosed — so you can judge whether our confidence scores mean what they say.

Every figure below answers one question: “When we said a home win, draw, or away win was most likely before kickoff, how often did that outcome actually happen?” Updated hourly from the last 180 days of finished matches.

Matches evaluated

62

Finished fixtures · 180 days

Overall result accuracy

45.2%

Top pre-match pick vs final score

Strong projections (≥70%)

51.3%

39 high-confidence matches

Avg. projection strength

76%

Mean confidence across sample

What these percentages refer to

  • Result accuracy — the share of matches where our single strongest pre-match outcome (home win, draw, or away win) matched the final whistle. We do not count near-misses or secondary picks.
  • Projection strength — how decisively the Monte Carlo engine favoured one outcome over the others before kickoff (0–100%). A 72% reading means roughly 72% of simulated scenarios landed on that outcome class — not a promise the result will occur.
  • Strong projections (≥70%) — fixtures where scenario separation was clear enough that we surface them on Pro Insights. These are the cases where our analytical stack found the most statistical edge.

When projection strength reaches 70% or higher

Across 39 matches where our model reported ≥70% projection strength, the top pre-match outcome matched the final result in 51.3% of cases — compared with 45.2% across the full 62-match sample.

Looking only at the 70–100% confidence bands in our calibration chart, combined result accuracy sits at 51.3% over 39 fixtures — evidence that clearer scenario separation in the simulation layer correlates with better real-world alignment.

Football remains stochastic; even a well-calibrated 75% projection misses roughly one in four outcomes. We publish these figures for research transparency, not as wagering advice.

Projection strength vs real outcomes

Matches are grouped by pre-match projection strength (50% and above). Expected accuracy is the average strength score the model assigned in that band; observed accuracy is how often the top-picked outcome (home win, draw, or away win) actually happened. When these bars align, the model is well calibrated — especially in the 70%+ bands where scenario separation is strongest.

Weekly accuracy trend

Share of matches each week where the strongest pre-match projection matched the final result. Short-term swings reflect sample size and fixture variance — not a streak to chase.

Accuracy by predicted outcome

When our model's strongest pre-match pick was a home win, draw, or away win — how often did that outcome occur? Draws are inherently harder to nail (lower base rates in most leagues); home and away picks in strong-projection fixtures tend to track closer to the 70%+ calibration bands above.

What goes into every projection

Before a confidence score is assigned, each fixture passes through our full analytical pipeline — the same signals you see on match pages (stats, form, standings, head-to-head). The accuracy figures above reflect how well that combined stack performed after the real goals were known.

  • ▸Expected goals (xG) and shot quality from rolling match windows
  • ▸Possession, shots on target, dangerous attacks, and corner pressure
  • ▸Home vs away scoring baselines calibrated per league
  • ▸Recent form weighted by opponent strength (not raw points)
  • ▸Head-to-head history over comparable sample periods
  • ▸League table position, points, and seasonal goal rates
  • ▸Squad market value and fixture congestion where data supports it
  • ▸Poisson–Monte Carlo outcome distributions (thousands of synthetic runs per fixture)
  • ▸Neural calibration layer that maps simulation bands to readable confidence scores

Full technical detail on the simulation engine, Poisson calibration, and neural confidence layer lives on our methodology page.

Accuracy by league

Competitions with at least 8 graded matches in the lookback window, sorted by sample size. Scoring culture and home advantage differ by league — so calibration varies, and thin samples can swing percentages.

No league meets the minimum sample threshold yet.

How we grade a projection

When a fixture kicks off, StatsBrain has already published a pre-match simulation: three outcome probabilities (home win, draw, away win) plus an overall projection strength score. After the match finishes, we archive the final score and compare it to what we said was most likely.

A hit means the outcome class with the highest pre-match probability — for example “away win” at 58% when home and draw were lower — matched the actual result. A miss means the football went another way. We grade every eligible fixture this way; there is no cherry-picking of markets, scorelines, or post-hoc “best of three” selections.

The calibration chart above tests a deeper claim: when we say projection strength is 80%, does the top pick win roughly 80% of the time over many matches? That is what separates a transparent research model from a black-box tipster — and it is why we publish this page publicly.

Honest limitations

A 75% projection still fails about one in four times — that is mathematics, not model failure. Small leagues and short weekly windows can move accuracy by double digits. StatsBrain is a research platform: we do not accept wagers or publish betting picks. Pair this dashboard with league context on scoring analysis, daily signals on Pro Insights, and live fixtures on the match board.