The why behind the numbers.

Each team carries four of those ratings: pass offense, run offense, pass defense, and run defense. The offense and defense ranks on the team ratings table come from a fixed blend of them, sixty percent pass and forty percent run, applied the same way on both sides of the ball. Net rating, the one figure the table prints outright, is a team's offense rating added to its defense rating. Positive is better everywhere, defense included, where a higher number means a unit that gives up less than an average one would.

That sixty-forty split is a modeling choice, not a count of snaps. Teams actually throw on between sixty-one and sixty-three percent of plays, and that share drifts from one season to the next. Holding the weight fixed means a rating carries the same meaning in an older season as it does in the most recent one, rather than moving with league-wide play calling.

The two teams' ratings are then set against each other. A strong passing offense meeting a weak pass defense projects differently than the same offense against a strong one. This is the matchup: not each team in isolation, but what each team's strengths and weaknesses become when they collide with this specific opponent.

The next question is how many plays the game will hold, because a projected efficiency only becomes points once you know how often it happens. Pace matters, and so does game script. A game expected to be close and competitive is played differently than one expected to be a blowout, where the team in front runs the clock and the game shortens. Touchdown Theory estimates the number of plays each team runs from the shape of the game it expects, so the play count is a product of the matchup, not a fixed assumption.

Efficiency and play count together produce the projected game: how many points each team scores, the total, the spread that falls out of the difference, and the probability each team wins. These are not four separate guesses. They are one consistent picture of the game, seen from four angles. The spread is the gap between the two projected scores. The total is their sum. The win probability is how often a game shaped like this ends with each side ahead. Because they come from the same source, they always agree with each other, which is a property the market's numbers do not always have.

This is the top of the chain. The game projection sets the stage the rest of the model plays out on: the environment it happens in, the players who fill the box score, and the odds each of them finds the end zone.

Wind is the biggest one. A strong crosswind or headwind pulls down the passing game and the kicking game, which lowers scoring and pulls a game's total toward the ground. Cold stiffens the ball and the hands and tends to suppress offense. A dome removes weather from the equation entirely and tends to sit among the best scoring environments in the league, which is why a fast track under a roof projects differently than the same two teams outdoors in December. Altitude thins the air in the one stadium high enough for it to matter, and the ball carries farther.

What makes this different from the way most coverage treats weather is that Touchdown Theory does not stop at noting it is windy. The environment is a measured adjustment inside the projection. The model was built by studying how games actually played out under these conditions across many seasons, so the size of each effect is learned from real outcomes, not asserted. When the total on a game moves because of a forecast, you can see how much of that move is the wind, the cold, or the roof, and separate it from the matchup itself.

Because it is a forecast, it firms as the game approaches. Early in the week a weather read is soft and the projection says so with a small marker rather than a false-confident number. By game day the forecast is reliable and the marker clears. That is the honest version of a weather-aware projection: the effect is real, it is measured, and its certainty is shown rather than hidden.

Most football coverage treats weather as a footnote, a mention that it is windy with no number attached. Touchdown Theory makes it a visible, measured part of the projection. That transparency is deliberate, and few products in football show their work this way.

The game projection produces team totals: how many yards a team is likely to throw for, how many it runs for, how many touchdowns it scores. The player layer takes those totals and hands them out among the players who will actually be on the field. Each player gets a share based on how he is used, his role in the offense measured through targets, carries, and snaps, combined with how efficiently he turns those chances into production. A featured back takes most of the carries, a number one receiver draws most of the targets, and the rest fill in behind them.

The property that holds the whole thing together is that the players add back up to the team. A team projected for two hundred fifty passing yards has its receivers' projected yards sum to that same two hundred fifty. Nothing at the player level ever contradicts the game above it. The chain reconciles from the game down to the last skill player, by construction, not by luck.

Injuries are handled by adapting, not hiding. When a starter is ruled out, his workload does not vanish, it redistributes to the players behind him, which is how the real game fills the gap. If a starting quarterback is out, the model shifts the usage and marks that the passing behind those numbers is now less certain, rather than showing the same confident line as if nothing had changed.

Behind each player's projection is a full range of outcomes, not just a single expected number. That range is what makes the next layer possible: the odds that each player finds the end zone.

It starts from the player layer above it. Each player already has an expected number of touchdowns for the game, built from his share of his team's rushing and receiving scores. From that expected number, the model produces the probability of at least one touchdown, the anytime-touchdown market that most people actually care about. A player expected to score often has a high probability of at least one, a player who rarely scores has a low one, and the model converts one into the other using the established mathematics of how independent scoring chances add up, not a rough rule of thumb.

That probability is then checked against reality. A raw model number is not trustworthy until it has been tested against what actually happened, so the touchdown probabilities are calibrated against years of real outcomes: when the model says twenty percent, it should happen about twenty percent of the time. That calibration is what separates a number you can lean on from one that just looks precise.

The result is a fair probability, and from a probability comes a fair price. A thirty percent chance to score has a fair price attached to it, and that fair price is what makes the next question possible: how does our number compare to the market's.

A fair value is what our model thinks an outcome deserves, shown as a probability with the fair odds beside it. When our probability is higher than the price the market implies, our model thinks that outcome is more likely than the market is pricing. When it is lower, the opposite. The gap between our number and the market's is the whole point of the comparison, and it is information, not instruction. Because we lead with the probability, our numbers are also directly comparable to prediction markets, not only sportsbooks.

This is the line the product holds on purpose. Touchdown Theory shows you the reasoning and the fair value. It does not tell you what to bet, it does not sell picks, and it keeps no record of wins and losses. What you do with the gap between our number and the market is your decision, made with a clear view of how our number was built and why it lands where it does. The value is in understanding the disagreement, not in being handed an answer.

That is the same principle the whole product runs on. Every number is traceable to the reasoning behind it. The fair value is simply that reasoning, expressed in the language of the market, so you can weigh it against what the market is charging and reach your own conclusion.

Football is genuinely uncertain, and no model changes that. A single injury, a bad snap, a ball that bounces the wrong way can decide a game that every input said should go the other way. Our projections are estimates of what should happen on average, and any single game can land far from the average. We hold the numbers to being logical and internally consistent, not to being right every time, because being right every time is not possible in this sport.

There are also things the way we build the model does not do, by design. Every number we produce comes from the same game projection, so they always agree with each other: the players sum to their team, the spread and total and win probability all come from the same projected score. What the model gives is the odds of each outcome on its own, the chance this player scores, the chance this game goes over. It does not produce the combined odds of several specific outcomes all landing together in the same game, which is what pricing a parlay correctly would require, and it will not flag a game as a statistical fluke. Products that do those things run a game thousands of times in simulation. We build each number from its parts instead, which is what makes the reasoning traceable, and the tradeoff is that a few simulation-only features are things we do not have. We would rather tell you that than fake it.

And the model is at its weakest exactly where football is least predictable. Early in the season, before there is much to learn from. Games with a late or uncertain injury picture. The preseason, which is the single hardest slate to project. Where the model is soft, the product says so, and the Accuracy page shows where it has been strong and where it has missed, openly, without hiding the misses.

None of this is a disclaimer bolted on at the end. It is the same idea the whole product runs on. A number you can trust is one that shows its reasoning and admits what it does not know.