The number starts from the market’s own numbers. Every posted line of the season so far is fitted to a rating for each team and one figure for home field, and the game being priced is left out of its own fit, so the market number for a game is what the rest of the season’s lines say it should be, not the line itself. The fit needs at least 32 lines; under that the market number is the posted line, the system holds no opinion, and every game is a lean. The market gap is the market number against the line, and it is stated on every pick as its own fact: the market’s own numbers make this line so many points too kind to one side.
Results are then allowed to speak, within limits. A second rating is fitted to the margins actually played, every margin capped at 21 points, blended toward a preseason prior from win totals at 0.85 points a win, with results weighted at played over played plus 6 and home field of 1.5 points. The form is half of the gap between that number and the market number, capped at 2 points either way, so results can move the market number and never replace it. Our spread is the market number plus the form, and the edge is our spread against the line. The form is printed on every pick as a fact beside the market gap; it is never an edge on its own.
Four signals move the edge by fixed amounts, and each keeps its own record above. Rest: a gap of 3 days or more between the two teams’ rest is 1.5 points to the rested side. Public: when 65% or more of the spread money is on one side, 1 point goes the other way. Weather: an outdoor game with wind at 20 mph or more, or a 60% or higher chance of rain, is 1 point to the underdog on the market line, and nothing on a pick’em. Quarterback: a team whose starting quarterback is out is docked 4 points, on live weeks only, because an injury status is a fact about now and not about a week that has already been played.
The gate. A game is a pick only when three things hold: the market gap is 3 points or more; at least one of the four signals fired on the same side as the gap; and the lean, the edge plus the signals, lands on that same side. A gap and one agreeing signal outweighed by the signals against and the form is a lean, never a pick the other way: a pick’s case never argues against its own side. Otherwise it is a lean. Every game gets a side, whichever team the lean favours, with a number of exactly zero going to the home side. Stars are bands on the size of that number: under 1.5 is one star, under 3 is two, under 4.5 is three, under 6 is four, and anything more is five; a pick is never fewer than three and a lean never more than two. Three stars and up are the Sharp picks and get a written case. One and two stars are leans, not picks: the side the number tilts to without clearing the gate, so they sit in the week table marked no play and are counted nowhere, not in the headline, not by stars, not in the backtest.
Over thirteen seasons the market-gap rule alone ran 51% against the close, and break-even is 52%. The ratings claim no edge. The signals are the case, and the closing-line block is where they are proved or not.
A pick is frozen the first time it is published with a line before kickoff: the side, the stars, the signals, the case and the line it was made against never change after that. A later run can update the current line, the score and the grade, and nothing else. The one exception is stated on the week it applies to, under the lead. A game that reaches kickoff without a line gets no row. When a live pick grades, the closing line is recorded beside it: the median spread across books from the last pull before kickoff, and how far the market moved toward our side between the pick and the close. Beating the close is the measure that survives one week’s scores, and it is kept in its own block above. Weeks 1 and 2 were run after the fact against the closing line and are labelled a backtest: a demonstration of the system, not a record.
The market ratings are the one thing fitted here, and they are fitted to the market’s lines, not to our results. Every other constant is stated so it can be argued with, and stays where it is until there are enough weeks to fit it honestly. The case on a pick and the lead on a week come from the Sharp Model, off the numbers on this page and nothing else; when the written case does not land, a numbers-only case stands in and says so. This page is a record of a system, kept in public. It is not a recommendation to do anything with it.