Read the
prices.
Every price is a claim about probability. We check the claim against our own model, built on the Sportsvyn power ratings and computed before we ever look at the price. When a number is generous, the market is paying more than the probability deserves. When it is rich, you are paying for certainty that is not there. We say which is which, and we keep score on ourselves below.
How grading works: a generous call lands when the outcome hits, because we said the probability was bigger than the price. A rich call lands when the outcome misses, because we said the price was overpaying. The third stat is the honesty check: at the prices we tagged, that is what the market itself expected to land. If our hit rate and that number converge over time, our reads are adding nothing, and this page will say so.
The market number. Prices are the median across 13 sportsbooks, refreshed hourly, with the bookmaker margin removed so the implied percentages sum to 100. What remains is the market probability, stated plainly.
The model number. Our probability is a Davidson three-outcome model driven by the Sportsvyn team power ratings alone. No form term, no host adjustment, no market input. The two parameters, k of 0.499 and nu of 1.153, were fit on 95 completed World Cup matches. It is our read of the game, computed before we ever look at the price.
The tags. Gap is our probability minus the market number. When the gap is 3.5 to 8 points in our favor the number is generous, and 3.5 to 8 points against us it is rich. Inside 3.5 points the price is fair and we leave it alone. One guard: above an 80 percent model probability we never call a number generous, because our calibration is unproven that high. The threshold is fixed and applies to every market on the board.
The wide state. Gaps beyond 8 points are disclosed as model-market disagreement, not value. At that distance our model is more often the limitation than the price, usually a stale or thin rating, so we flag the gap and trust it less, rather than dress it up as an edge.
What we cover. Tags appear only where our model prices the market. The match result carries the Davidson model. Our totals model is calibrated at the tail lines (over/under 1.5 and 3.5), proven walk-forward; the main line is shown as market information because the model has no demonstrated edge there. Tail lines appear on the board only when the model has a read. Scorer prices are single-sided and shown as offered, so they still carry the book margin. Model coverage expands only by a stated projection and calibration, never by default.
What this is not. A generous tag is a statement about a price, not a prediction of an outcome. Most generous underdogs still lose; the claim is that they win more often than the price pays. Nothing on this page tells you what to do, sizes a stake, or links to a book. We publish the gaps and the ledger, and the ledger grades us.