Know before you pick

What actually sits behind a 58% probability: Elo, Poisson, Dixon-Coles, boosting, calibration — and the honest limits of every one of them. 18+

Why a probability on its own is not enough, what analysis strength measures, and how it raises the bar a fixture must clear before it becomes actionable. 18+

Why hit rate is a weak measure, what it means for a 60% forecast to be honest, and how Brier score and log-loss punish confident wrong answers. 18+

The three classical models behind most football probabilities: what each measures, where each fails, and the parameters that decide it. 18+

62% probability with a confidence score of 40 is not a contradiction. What each number answers, why both are published, and how it looks on a real sealed record. 18+

How to convert probability to a fair price, strip bookmaker margin two different ways, and measure an edge in percentage points instead of nonsense. 18+

How to compute EV, how fast a small estimation error erases it, and why closing line value grades a decision without the result. 18+

The formula, a worked example, and why full Kelly is almost always wrong when your probability is an estimate rather than a fact. 18+

Most fixtures do not deserve a decision. The thirteen decision states, the uncertainty budget, and why abstention is the habit that pays off fastest. 18+

Why a winning result proves nothing, the four things you can grade before the ball is kicked, and how a decision-quality score is built. 18+

Why percentage staking beats fixed staking, the early warning signs of chasing, and which barriers hold when judgment is compromised. 18+

Why multiplying odds assumes something false, the four kinds of correlation in football slips, and how embedded margin grows with every leg. 18+

Identifier, timestamp, SHA-256 over a canonical payload and an append-only ledger — how anyone can check that a prediction was not rewritten. 18+

What a good tipster does better, what a calibrated model does better, and the one test that separates a product from a content schedule. 18+

Seven questions that separate a real model from a nice interface with a number on it — plus a two-week test that costs nothing. 18+

Seven checks before you trust any prediction source: when the record was written, whether losses are shown, the price at entry, and self-verification. 18+

The league with the best data and the tightest prices. What that means for edges, and where a model still adds value. 18+

How ratings and attack/defence ratios are computed per league, why cross-league comparison is invalid, and what that changes. 18+

Why the draw is the hardest outcome to forecast, how a correction to four scoreline cells moves it, and what a draw radar can and cannot tell you. 18+

How expected scoring rates become a matrix, why every market is read off the same object, and when goals markets are withheld entirely. 18+

Why a five-match form line measures the fixture list, what opponent-adjusted form fixes, and what the model will not encode. 18+

Ratings live inside closed systems, so European ties need a bridge. What that costs in confidence, and what the backtest omits. 18+