Anyone can claim a model works. This page is the actual test PitchLine holds itself to — run against real results, scored against a simple benchmark, with nothing left out of the average.
Two numbers, both standard in forecasting — not invented for this page.
Brier score
The squared error between the probability PitchLine gave an outcome and what actually happened. A confident, correct call scores near 0; a confident, wrong call is punished hard. Lower is better. Guessing every outcome as equally likely scores about 0.667 — that's the floor a real model has to beat.
Benchmark: typical rates
A deliberately simple rival: for every match, predict the home, draw and away rates seen in that league so far. PitchLine only adds something if it scores better than this. Here the benchmark scores about 0.64, well below the 0.667 of treating every outcome as equally likely, so it is the fairer bar.
Versus real bookmaker odds — not measured yet
The harder test is to score the real closing bookmaker odds on the same matches and compare. Closing lines are shaped by professional money and are genuinely hard to beat; expect them to be sharper than a simple model like this one. This comparison needs odds data PitchLine does not have permission to use yet, so it is not reported.
Walk the season in order
Every match is predicted using only results that happened before it — never data from later in the season. A model that peeked at the future would look better than it is.
Score every prediction, not just the highlights
The full season's matches go into the average — not a curated set of the calls that went well.
Compare against a simple benchmark
Every prediction is also scored against a naive "typical rates" forecast on the same matches. A comparison with real bookmaker closing odds is planned, once there is data PitchLine is allowed to use for it.
A backtest across the Premier League, La Liga, Bundesliga, Serie A and Ligue 1: last season (2025-26) plus the current season (2026-27) up to the latest data available. Lower scores are better.
Across 1,757 predictions, PitchLine scored 0.599 against 0.636 for the "typical rates" benchmark, and picked the most likely outcome correctly 52.1% of the time.
Whole period, by league
| League | Predictions | PitchLine | Benchmark | Gap | Top pick |
|---|---|---|---|---|---|
| Premier League | 375 | 0.618 | 0.650 | −0.032 | 49.3% |
| La Liga | 406 | 0.584 | 0.622 | −0.038 | 52.5% |
| Bundesliga | 304 | 0.587 | 0.640 | −0.052 | 54.6% |
| Serie A | 371 | 0.598 | 0.638 | −0.039 | 52.3% |
| Ligue 1 | 301 | 0.608 | 0.631 | −0.023 | 52.2% |
| All five | 1,757 | 0.599 | 0.636 | −0.037 | 52.1% |
Gap = PitchLine minus benchmark; negative means PitchLine scored better.
Current season only (2026-27 so far)
| League | Predictions | PitchLine | Benchmark | Gap | Top pick |
|---|---|---|---|---|---|
| Premier League | 42 | 0.618 | 0.674 | −0.056 | 47.6% |
| La Liga | 61 | 0.553 | 0.621 | −0.068 | 54.1% |
| Bundesliga | 30 | 0.492 | 0.591 | −0.100 | 60.0% |
| Serie A | 47 | 0.569 | 0.635 | −0.066 | 57.4% |
| Ligue 1 | 39 | 0.630 | 0.638 | −0.008 | 53.8% |
Samples this small (30 to 61 matches per league) swing a lot from week to week. Treat these as an early look, not a verdict; Ligue 1's gap in particular is too small to tell apart from chance.
What this does and doesn't show
A backtest, not a live record
Each prediction was computed afterwards, using only matches played before it. It was not logged before kickoff. A live record, with predictions published ahead of each match, starts once PitchLine publishes them in advance.
The bar is low on purpose
Beating "typical rates" is the minimum a model has to do. The real test is the bookmakers' own odds, which are not measured here yet, and closing lines are usually sharper than a simple model like this.
The data isn't perfect
Results come from a volunteer-maintained open dataset, and it is incomplete: for example, 353 of 380 Premier League matches from last season have scores, and the current season only runs up to the latest matches the dataset contains. These numbers reflect only the matches it held.
Nothing was tuned to these results
The model's one adjustable setting, how strongly a team's rating is pulled toward average until it has played enough games, was set using simulated data before this run and has not been changed to improve these numbers.
A second, more concrete way to see whether the model is actually finding value — not just whether it sounds confident.
How it works
A fictional $10,000 bankroll, flat $100 stakes, betting only when PitchLine's probability for an outcome is meaningfully higher than what the real market's odds imply — no bet at all when the model and the market roughly agree. Every result runs through the same rolling backtest as the accuracy score above: only past data is ever used to predict a match, and every bet is graded against real market odds, not PitchLine's own.
When tracking starts
Fictional stakes only begin once a covered league's current season actually kicks off — never backdated into a prior season to make the numbers look better. The model still uses the prior season's results to build sensible team ratings beforehand (so it isn't guessing blind on opening weekend), it just never bets on any of that earlier data.
What this isn't
This is a hypothetical simulation for illustration only — not real money, not a real betting account, and not gambling or investment advice. A run of good weeks here would not be a promise about future weeks; sports betting carries real financial risk, and a fictional bankroll doesn't change that.
The simulator is built and tested, but it deliberately refuses to run without real bookmaker odds, because a hypothetical profit has to be measured against a real market. The open data PitchLine currently uses has no odds, so this stays empty until a source with odds can be used. When it runs, the full daily breakdown will post here, wins and losses both.