How we rate games

What each number on Research means, where it comes from, and when we hold one back.

The names

Portal line: Our model’s fair numbers for an NFL game: a fair spread, a fair total and a fair win chance. The AI scout writes them each day from the game’s research packet.
AI read: The written report behind a game: what the scout saw and why it leans the way it does.
Power rating: On college games, ESPN’s power rating blended with ratings built from this season’s betting lines, plus home field. It is context for the AI read, not a Portal line.
Your line: The line you build yourself from the factors on a game page.

NFL edges

Each market compares the Portal line with the current market line. Spread and total edges are in points. The moneyline edge compares two win chances: ours, and the one the market prices in once the bookmaker’s margin is taken out, shown as “53% by our read vs 46% priced in”.

A market counts as an edge once it reaches 1 pt on the spread, 1.5 pts on the total, or 3 points of win chance on the moneyline. When a game has more than one, the market that clears its minimum by the most leads. The board ranks games the same way, then by confidence (slight, solid, strong), then by kickoff.

Check this

Some numbers are too far from everything else to trust without a look. We mark a market Check this, say why under its call, and leave it off the ranking when:

a fair number sits more than 1 point outside the range of its sources (the market, nfelo and The Power Rank on the spread; the posted total, the opening total and the scoring baseline on the total);
the edge is over 4 points on the spread or 6 on the total;
on the moneyline, our win chance is more than 8 points from both the market and ESPN’s matchup predictor, favors the other team from our own fair spread, sits on the other team from the spread edge, or is more than 5 points away from what that spread is worth.

A read whose note argues the other side of its own number, or whose three numbers disagree with each other, is not flagged but held: it shows as rechecking and reruns.

The record

Every market the Portal line took a side on is graded once the game is final, at the closing line: the spread and the total flat at -110, the moneyline at the closing price. A side counts only when it still cleared its minimum at the close. The record under the board starts with Week 5, shows each market and whether the number beat the close (measured from the market when the read first went up), and keeps flagged plays apart from ranked ones.

Reruns

A read reruns on its own when the line moves 1.5 points or a starting quarterback changes, and when its numbers disagree with each other. While that happens the game shows as rechecking and leaves the Edges list; it stays under All games. Members can also rerun a game themselves; a rerun stays until kickoff unless the line or injuries change.

College

On college games the AI read leads. The power rating sits under it as context. A college edge needs 3 points on the spread, 4 on the total, or 5 points of win chance on the moneyline, because the college model is lighter than the NFL one. The same Check this rules apply, with the power rating and its two parts (ESPN’s FPI and ratings fit to this season’s closing lines) as the spread’s sources, and a read is flagged when either team is short of FPI or 3 games with a closing line.

Player props

Every posted prop starts on the market’s number: the books’ consensus line, read at their no-vig price, with the margin taken out of a one-sided market such as an anytime touchdown. The AI scout moves a number only for a reason it can cite from the brief (an injury, a usage change, the matchup, weather, the game script); the move is capped at one game’s worth of spread and scaled by its confidence. That number becomes a chance the prop goes over, and the edge is that chance against what the best price across books has to beat, in points of win chance.

A prop is an Edge at 5 points (8 on touchdown props), a Lean at 3 (5), and on Watch at 0.5 (1); a one-sided market needs 2 more in every tier. Badges cost points instead of removing a prop: a questionable or limited player, a questionable quarterback, a thin market, stale usage data, a price far longer than the other books. Half of the best price’s premium over the other books comes off too, so a book that is simply off the others cannot make an edge on its own. A player listed out or doubtful, a line that has not refreshed, a prop under two books, and a note that argues the other side are left out.

Contingent rows price a teammate as if a questionable player sits. Market only rows are a book beating the no-vig consensus by 2.5 points with no projection needed. The record under the tab counts every Edge and Lean as it stood at kickoff from Week 8, graded from the box score and against the closing line.

AI leans

The AI leans tab shows the picks of the AI players in the weekly contest. They read the same research and make their own calls, so they can disagree with the Portal line.