Chapter 6 · 6 min read
Chapter 6 — Glossary: xG, Glicko-2, Brier, edge, Kelly and every key term
Every time you read a PronoStats prediction — or any other AI model — you bump into these terms. Knowing what they mean is the difference between executing a decision and trusting a buzzword. Bookmark this chapter: come back whenever you hit an acronym you don't remember.
xG (Expected Goals)
Probability that each chance becomes a goal, based on shot location, angle, body part, defensive pressure. A long-range shot is worth 0.03 xG, a penalty 0.78. Sum match xG and you get how many goals the team 'deserved' to score. Long-term, goals converge to xG: if a team scores much less, it's statistically unlucky.
Glicko-2 and Brier score
*Glicko-2 is a rating like ELO but with two additions: a deviation (RD = how sure we are of the rating) and a volatility. A newly-promoted team has a high rating but huge RD = the model knows it doesn't know. Brier score* measures how far predicted probabilities are from real outcomes. 0 = perfect, 1 = terrible. Under 0.25 for 1X2 is great. PronoStats sits around 0.79 raw, 0.64 after Groq correction.
Edge, EV (Expected Value) and value bet
*Edge = (your probability × odds) - 1. Measures how much better your estimate is than the bookmaker's. Positive edge = potential value bet. EV* is the monetary expectation per unit staked: stake × edge. Example: stake €10, edge 10% → EV +€1. Doesn't mean you win €1 tonight, it means IF you play 1000 identical bets, you average +€1000. Single bets can lose; the portfolio wins over time.
Kelly criterion and fractional Kelly
Formula to size your stake: f = (b·p - q) / b, where b = odds - 1, p = your probability, q = 1 - p. Returns the optimal fraction of bankroll to risk. Mathematically maximizes geometric growth rate. *Problem: psychologically brutal (in a losing streak you can lose 60% in 10 bets). Solution: ½ Kelly (calmer) or ¼ Kelly* (much calmer). Pros use ¼.
ROI, drawdown, yield
*ROI (Return On Investment) = profit / capital invested. A season with +€200 on €2000 staked = +10% ROI. Yield is the same for bettors = profit / stake volume. Good range = +3-7% long-term (above 10% = suspect or small sample). Drawdown* = max loss from peak to next trough. Tracking drawdown is even more important than ROI: it tells you if your strategy lets you survive to the end.
Markets: 1X2, BTTS, DNB, AH, Over/Under, push
*1X2 = home / draw / away. BTTS (Both Teams To Score) = do both teams score? Yes/No. DNB (Draw No Bet) = draw refunds. AH (Asian Handicap) = handicap with fractions that eliminate the draw. Over/Under = total goals above/below a line (most common: 2.5). Push* = when the outcome lands exactly on the handicap or line (e.g. 1-1 on AH 0) and the bet refunds the stake — no win, no loss.
Chapter quiz
Check if the concepts stuck — nothing tracked, just for you.
1.Edge = +12% on odds of 1.85. What does it mean?
2.Brier score of 0.30. How do you interpret it?
3.On Asian Handicap AH 0, the score is 1-1. What happens?
Frequently asked questions
Is xG reliable for all leagues?
Best in top leagues (Big 5) where xG models train on huge samples. In minor leagues the data exists but with higher uncertainty.
What's a 'realistic' yield for a part-time bettor?
+3% to +7% long-term is excellent. Sustained above 10% over 500+ bets is rare. Below +2% you struggle to beat variance.
Can I trust a model with a small sample?
No — under 50 bets for 1X2, 30 for Over/Under, you're a victim of variance, not the model's truth.