An NFL model that shows its work.

Plain-English research: what the model likes, where the best number is, and whether it is actually working.

This week

The calls

Each card is one player prop the model flagged, with the honest case on both sides and how hard it leans. These are research notes, not tips.

The board

The best number, everywhere.

The best price on each side across the books we track, right now. Taking the best available number is the one edge that needs no prediction: the same result simply pays more.

Every moneyline shows each team's plain-English chance to win the game outright, from the sharp price with the vig removed. Where a book beats the sharp reference (the price-setting books other lines follow) we mark it ◆ better with the estimated edge, and where two books' numbers overlap so one final score can cash both sides of a bet we flag a ◆ middle and show how often it lands. All of it is research to look closer at, not tips. Lines move fast, so confirm the number is still live at the book.
Track record

The honest scoreboard.

This is a paper record: calls are logged and graded, but no real money is placed. The point is to find out whether the model beats the market before anyone risks a dollar.

CLV is short for closing line value. In plain terms: after the model flags a number, did the line move toward the model or away from it by the time the game kicked off. If the line keeps moving your way, you got a good price. It is the cleanest signal of a real edge because it ignores whether one game happened to win or lose.
For research and entertainment. Not betting advice. Past results do not guarantee future outcomes. This is a paper record, not a wagering record. 21+. If gambling is a problem, call 1-800-GAMBLER.
How it works

What this is, and what it will not pretend.

nickspicks runs on one idea: be honest about where an edge actually is, and label everything else for what it is.

The single guaranteed edge is price: taking the best number across books on anything you follow. That is arithmetic, not prediction. A better price pays more on the exact same outcome. Everything past that is either a short list of angles that have held up in testing, or context that only helps you think.

CLV (closing line value) is the scorekeeper. Beating the number that closes is what proves a price was real once you strip out the luck of any single game. Win-loss records swing wildly over a season. CLV is steadier and harder to fool.

What actually counts as an edge

Ranked by how much to trust it. Only the tested items are treated as more than context.

Best price Guaranteed
Taking the best available number on everything you follow. The same outcome simply pays more. Always on the table.
Receiving-yards regression Tested
The model reads receiving-yards lines that the opener has not yet caught up to, often because a prior-week injury or role change has not been priced. It has held up across two seasons of paper testing. Still paper only, on the open, on retail books.
Injury and role-change timing Watching
Catching a line before it adjusts to a snap-count or depth-chart change can have value, but only live, and it is not proven. Treated as a lead, not an edge.
Everything else Context
Game reads, projections, power ratings. Useful for thinking, not for betting. The biggest model-versus-market gaps are usually the model missing something, not finding something.

The model, the context, and the smart layer

Three layers do three different jobs. Each hands off to the next, and the last one is not allowed to make anything up.

The model does the math
Power ratings and projections turn into a number for every game and every player prop, then compare that number to what the books are posting. Where the two disagree by enough, it gets flagged. This layer is pure calculation, no opinions.
Output: a projection and a flag
The context layer gathers the facts
For each flagged prop it assembles a fact sheet of everything we can actually measure: the line and our projection, recent usage and target share, the matchup, coverage tendencies, and any prior-week injury or role change. Only numbers the code computed make the sheet, nothing hand-waved.
Output: a sheet of checked facts
The AI-smart layer writes the read
A language model turns that fact sheet into a short, plain brief: the case for the bet, the case against it, a verdict, and a one-to-five conviction. It is told to argue both sides honestly and to say pass when the facts do not justify a bet.
Output: the card you read on This Week
The guardrail that ties them together
The smart layer may only use numbers that appear on the fact sheet. Every card is checked after it is written, and any card that cites a number the model did not compute is thrown out automatically. The AI makes the research readable, but it is never allowed to invent a statistic to talk you into a bet.
Why you can trust the card

Team power ratings

One number per team, in points, where zero is a league-average team. The gap between two teams is the model's spread on a neutral field: if one side rates +6 and the other +1, the model makes the first team five points better before home field.

The rating blends three independent measures, each tested on its own before they were weighted together:

Efficiency (EPA)
Points earned and allowed per play. The sharpest read on how good a team actually is right now, week to week.
Track record (Elo)
A running rating that rewards beating good teams and moves with every result. Slow to fool, good at the long view.
Margin and schedule (SRS)
Average scoring margin, adjusted for how tough each opponent was. Keeps a hot start against weak teams from being overrated.

Before the season the ratings are seeded from Vegas win totals rather than a blank slate, then updated every week as games are played. A separate quarterback adjustment nudges a team up or down for who is actually starting. It is context to help you think, not a set of bets.

For research and entertainment. Not betting advice. Past results do not guarantee future outcomes. 21+. If gambling is a problem, call 1-800-GAMBLER.