VectorPicks is a numbers-first NFL opinion project. Vector, our AI analyst, builds its own point spreads, compares them with the line we can actually get, and labels a pick only when the evidence clears a defined bar. This page explains the current v2 method in plain English.
1. Team power ratings
Each team gets a power rating in points versus an average team on a neutral field. A positive number means better than average; a negative number means worse. The rating is a point estimate, not a rank, so the difference between two teams can be turned into a spread.
For the preseason starting point, v2 uses the 2026 sportsbook win totals from DraftKings, reported by FOX Sports in September 2026. We translate the market’s expected wins into points at 1.5 points per win above league average. This gives us a reasonable starting point before enough games have been played. The starting point is split evenly between offense and defense because the win-total market does not give us a sourced offense/defense breakdown.
As games are played, the starting point gradually gives way to the results. The results weight is:
That means early games matter, but they do not make a team jump wildly after one unusual result.
2. Results and efficiency
The results portion is an even blend of two views of performance:
- Final-score margin: points scored minus points allowed, adjusted for the opponent.
- Play-by-play efficiency: opponent-adjusted expected points added (EPA), which looks at the value of individual plays rather than only the final score.
Each view supplies half of the results component. The combination keeps real scoreboard outcomes in the model while giving efficiency information a seat at the table. Ratings are re-centered so that the league average remains zero.
3. The player-value injury layer
Team ratings describe a healthy baseline. We then apply a separate, visible player-value layer for verified availability news. The value is our estimate of the difference between the player and the replacement, not a claim that every injury can be measured perfectly.
- Quarterbacks have the largest values because changing the quarterback can move a team more than most single-player absences.
- Other values depend on the player’s role and the quality of the replacement. We identify uncertain statuses rather than pretending they are settled.
- We show the adjustment so readers can see what changed from the healthy rating to the game-day number.
Injuries are not a reason to chase a stale edge. If the available line has already moved by the value of the news, the game can be a Pass even when our adjusted number still differs.
4. Fair spread
- Home field is 2.5 points by default and 0 at a neutral site.
- Rest and travel are included only when we can verify them. We do not fill gaps with guesses.
- We compare this fair spread with the line we can actually get and publish. We do not compare against a better number that appeared later or an imaginary line.
5. Edge, checklist and tiers
The edge is the difference between our fair spread and the available line, measured toward the side we prefer. A number alone is not enough for a Best Bet. V2 uses four checklist signals:
- V2 edge: our v2 number shows an edge of at least 2 points.
- V1 agreement: the older v1, score-based model also shows an edge of at least 2 points.
- Market confirmation: the market action or current available line supports the side rather than contradicting it.
- Verified injury or matchup edge: sourced player availability or a specific, verified matchup supports the side.
| Tier | Rule |
|---|---|
| Best Bet | Our v2 edge is at least 2 points and at least 3 of the 4 checklist signals hit. |
| Lean | A smaller edge or a promising but incomplete case that is worth tracking without calling it a Best Bet. |
| Pass | Anything else: no usable edge, insufficient support, or a line that has already absorbed the relevant news. |
We keep the language proportional to the evidence. A Lean is smaller; a Pass means we are not claiming a playable edge. The line used for comparison is the one we can actually get, not a hypothetical better price.
6. Outside opinions
We may show other handicappers’ selections or AI-model opinions for reference and context. They are not part of our rating calculation, never count as checklist hits, and never create a Best Bet by themselves.
7. Trends
Historical trends can be interesting, but they are noisy. We may mention a trend in a footnote for color; trends never count toward a pick or checklist. Verified rest and travel belong in the fair-spread calculation, not in a retrofitted trend argument.
8. Integrity and timing
- Method change: we changed to v2 on September 25, 2026, before Week 3 Sunday.
- Thursday honesty: Thursday’s graded picks stay graded exactly as published. We do not rewrite them using the new method.
- No invented numbers: missing lines, injuries, splits, or statistics are shown as missing rather than guessed.
- Losses stay visible: graded picks remain on the record, wins and losses alike.
- Entertainment and opinion only: nothing here is betting advice. 21+.