Public track record
Every analysis is kept, winners and losers alike. The figures below are computed only from real, settled analyses.
Analyses are for information only and do not guarantee any result.
Model accuracy
How good our probabilities are, match by match
Computed on every match analysed and finished, whether or not a bet was tracked. Nothing is selected after the fact.
| Favourite's probability | Matches | Stated | Observed |
|---|---|---|---|
| < 45 % | 62 | 40,4 % | 43,5 % |
| 45-55 % | 50 | 48,9 % | 52,0 % |
| 55-65 % | 46 | 59,1 % | 56,5 % |
| ≥ 65 % | 36 | 74,0 % | 88,9 % |
Good accuracy does not make a good bet: the odds already include that probability. That is why a bet is only tracked when the price beats the estimate. Small samples: the gaps narrow as volume grows.
No bet has passed our value filters yet.
The engine only recommends a bet when the best market price beats the estimated probability by at least 2.5%. In our backtests that applies to 1 to 2% of matches, roughly one bet a week across five leagues. Nothing is simulated: the model's accuracy on every finished match is published above.
Back to the dashboardBacktest
What the method is worth over three seasons and five leagues
Walk-forward test with no information leak: on each matchday only earlier data is used. Log loss measures the quality of the probabilities (lower is better). The betting rule tested is exactly the one used on the site: best price at least 2.5% above the fair price, 1X2 and over/under 2.5 goals.
161 tracked bets over 4,806 matches · yield +9.7% · average CLV +2.8% · +15.6 u
An honest reading: the observed yield (+9.7% over 161 bets) carries a lot of uncertainty, and the last season is negative. The most reliable indicator is the average CLV (+2.8%): the prices taken beat the closing prices. Bet volume falls year after year as bookmaker prices converge. The best price comes from a multi-bookmaker feed and is not always available in your country or on your platform: always compare with the fair odds. Pronos AI is an analysis tool, not a guarantee of profit.
Counting method
How the figures are computed
Simple rules, fixed in advance and applied to every analysis.
- 1
1 unit per bet
Each tracked analysis counts as a one-unit stake, whatever the confidence. This avoids flattering the results by weighting after the fact.
- 2
Profit = odds − 1
A winning bet returns odds − 1 unit (odds of 2.10 return +1.10 u). A losing bet costs −1 u.
- 3
ROI = profit / decided bets
Return on investment divides total profit by the number of bets won or lost, i.e. the number of units actually staked and decided.
- 4
Voids excluded
A void bet (postponed match, voided market) is worth 0 u. It counts as a settled analysis but enters neither the win rate nor the ROI.
Analyses in progress (not settled) do not appear in any indicator until the match is over.