Here is how the projections are actually built.
We run the same method across the NFL, the NBA, and MLB, and we will show you the honest, category-level version of it. This page names what feeds a projection, explains the discipline that freezes it, and describes how every number gets checked against what really happened. We do not hide behind a black box, and we do not hand over the formulas either.
What feeds a projection
Every number starts from the same families of information, whatever the sport. We name the categories here and give an example from each league, but we do not publish the weights, formulas, or thresholds that combine them, because that part is the model itself.
Official stats and current form
We start from player and team performance in the official record, season-long numbers next to what someone is producing right now. A running back's recent workload, a guard's scoring over the past two weeks, and a starting pitcher's season line all live in this category.
Matchup and role
This category asks who faces whom, and what job that player actually has. In the NFL it is a back's work near the goal line, in the NBA it is a player's minutes and usage in the rotation, and in MLB it is a pitcher against the specific lineup and its handedness.
Venue and conditions
Where a game is played changes what happens in it. Wind and cold move an outdoor NFL game, travel and rest shape an NBA night, and ballpark dimensions with game-time weather push MLB run and home-run environments up or down.
Quality and opportunity signals
We look at how production is generated rather than at the box score alone. That means where an NFL player's carries and targets come from, how a shot or a possession is created in the NBA, and Statcast-style contact and pitch quality in MLB.
Live market prices
We pull current sportsbook odds across every market we cover. The market price is the benchmark a projection gets measured against, and the gap between the two is what surfaces edge.
Confirmed lineups and inactives
The last input to settle is who is actually playing: the NFL inactive list, the NBA starters and rest decisions, and the MLB lineup card with its probable pitchers. That is the point at which a projection is treated as final.
Final once the lineups are confirmed
A projection stays a living number for most of the day. It moves as inactive lists post, as starters and rest decisions are announced, as weather updates land, and as market prices travel. The estimate you see always reflects the latest information rather than a stale guess from that morning.
Then it stops. Once the day's lineups or inactives are confirmed, a projection is treated as final for that game, its inputs are settled, and the number is frozen. That freeze is what makes honest grading possible, because a projection cannot be quietly changed after the fact. What you saw before kickoff, tip-off, or first pitch is what gets scored.
The projection versus the market
A projection on its own tells you only half the story. The other half is the market price, which is what sportsbooks are charging for the same outcome. We convert that price into its fair probability by removing the built-in margin, then we set it beside the model's own number.
The gap between the two is what we call edge. When the model rates an outcome more likely than the fair price implies, the edge is positive. Edge measures relative value rather than certainty, and it never guarantees a result. Every board is built to surface that gap and label it with a confidence tier, so the biggest disagreements with the market rise to the top. The method reads the same way on an NFL touchdown prop, an NBA points line, and an MLB run total.
Graded against real results, every day
Frozen projections are compared to what actually happened, day after day and sport after sport. That daily grading is how the approach gets refined over time, and it is how we can tell whether a tool is doing its job. We keep an actual record against reality rather than a feeling about it.
It also sets a rule we hold ourselves to. A model is only marketed with a track record once that record has been independently validated. Until then we will tell you what a tool does and how it reasons, which means the categories above, the freeze discipline, and the edge comparison, without dressing any of it up in performance claims it has not earned. That restraint is the whole point.
How the models work, answered
What kind of data feeds these projections?
The inputs fall into a few broad categories: official player and team performance, matchup and role, venue and conditions, quality and efficiency signals, live market prices, and the confirmed lineups or inactives for that day. The model blends those categories into one expected number for each outcome.
Does the same method apply to the NFL, the NBA, and MLB?
Yes. The input categories and the discipline are shared across sports, and only the illustrations change. Role and usage carry an NFL projection, minutes and pace carry an NBA one, and the pitcher against the lineup carries an MLB one. In every case an independent projection is compared to the market price.
Does the model use real-time data like lineups, inactives, and odds?
Yes. Projections refresh through the day as inactive lists, confirmed lineups, late scratches, updated weather, and moving market prices arrive, so the number you see reflects the latest available information rather than a static estimate written that morning.
How often are projections updated?
They update repeatedly through the day and are treated as final once that day's lineups or inactives are confirmed. Until then a projection can move with new information, and after that its inputs are settled for the game.
Does a human review the output?
The model produces the projections, and the results are tracked and reviewed against real outcomes every day. That review is how the approach gets refined over time and how the discipline behind each tool stays honest.
How do you know if a model is any good?
Output is graded against actual results every day, and a model is only marketed with a track record once that record has been independently validated. We describe what a tool does and how it reasons before we make any performance claim.
See the method on today's board
You get independent projections next to the market price, the edge called out, and a confidence tier on every pick. Game lines and player props get the same treatment in every sport we cover, with the NFL and NBA boards arriving as those seasons open.