NFL Monte Carlo Sim

NFL picks and predictions this week

This tool simulates every NFL game on the week's schedule 20,000 times at the drive level and reports the full range of outcomes rather than a single projected score. For each game you get a win probability for both sides, an over and under probability against the posted total, a spread cover probability, and the gap between each of those and the price the sportsbook is offering. It is free with an account.

The simulation runs on the server once per day through the season, not in your browser, so the numbers on the page are the same numbers the model published when it made its picks.

Why the simulation works at the drive level

Football margins are not smooth. They pile up on 3 and 7 because scores are built out of field goals and touchdowns, so any model that draws a margin from a smooth bell curve misprices a spread sitting on a key number.

Each iteration draws a possession count for both teams, resolves every drive as a touchdown, a field goal or nothing at rates solved to hit that team's expected points, then plays a score-aware final possession. That endgame layer is what puts real weight on the key numbers: a team down 1 to 3 plays for the field goal, a team down 4 to 8 has to hunt the touchdown because a field goal is worthless, and the team in front runs clock. The key numbers fall out of the scoring units rather than being pasted on afterwards.

Why the model does not read the game line

Expected points are rebuilt from team season ratings and adjusted for weather, not taken from the posted line. Anchoring a simulation to the market and then comparing the output to the market would be comparing the market to itself, which produces an edge that looks real and is not.

The published probability is then blended 80 percent market and 20 percent model, and a pick is only made when the blended probability clears the price by a set margin. That blend is deliberate: the market is a strong baseline and the simulation is a second opinion about where to look closer.

Totals and moneylines are published, spreads are not

Measured against the 2025 season the simulated margin distribution is about 5 percent too narrow, which makes it overconfident on any spread away from the key number. Spreads are still simulated and still graded, they are simply not published as bets until that distribution is fitted. Totals and moneylines do not depend on the key numbers and are published.

Picks are only written inside three days of kickoff. An NFL line moves all week and a pick is recorded once at the price it was made, so an opening-line pick would permanently block the pick the model would make at the number you can actually bet.

What the History tab is for

Every published pick is graded against the real final score, pulled from the league data feed after each game window. The History tab shows the running record split by bet type, a cumulative profit and loss line on a flat one unit stake, and a breakdown by how big the edge was when the pick was made. Break even at standard -110 juice is 52.4 percent, so that is the line any record has to clear before it means anything.

Frequently asked questions