What each game is worth. Pup
Each bar runs from a team's chance to make the playoffs if it loses its this week's game to its chance if it wins. The tick is where it stands now. The longer the bar, the more that one result is worth, and games are ordered by how much they swing both teams.
Each bar: a team's playoff chances if it loses → if it wins this week. The tick is now. Longest bars first.
Simulation output, rebuilt daily from real results and nfelo game probabilities. Ungraded, and not betting advice.
This week's games
Flip a result to see where both teams land. One game at a time: every other game stays simulated. Hover or focus a row for the exact numbers.
Playoff odds now, all 32
Share of simulated seasons in which each team makes the field, wins its division or takes the #1 seed and its bye. Records are real results; the rest is simulation.
Method — how these numbers are made, and where they can be wrong
The simulation
Every day after the nfelo refresh, 100,000 copies of the rest of the season are played. Finished games keep their real result (a tie counts as half a win, as the league counts it). Every unplayed game is one coin flip weighted by its home-win probability from /data/survivor.json: the published nfelo forecast for this week, the nfelo season projection after it. Then each conference is seeded the league's way — four division winners and three wild cards.
What the bar measures
For each game this week, every simulated season is played twice: once with the home team winning and once with the away team winning, while every other game keeps the result it got in that season. "If they win" and "if they lose" are the share of those seasons in which the team makes the playoffs; "now" is the share with nothing forced; the swing is the difference. Because the two halves share every other result, the swing isolates what this one game does. The ± in the tooltip and table is the simulation's own error on the swing.
Tie-breakers
The NFL's division and wild-card procedures, in order: head-to-head (for three or more wild-card clubs, a sweep only), division record, common games (wild card: at least four), conference record, strength of victory, strength of schedule. A tie still standing after that would need points, which are not simulated, so it goes to a coin toss. The seeding code is checked against history: replaying the 2002–2025 regular seasons through it reproduces every conference's actual playoff field, byes and wild-card pairings, 48 of 48 conference-seasons, and the check reruns on every change to it.
Where it can be wrong
- Loading…
That is why the page prints <1 and >99 instead of 0 and 100: 100,000 seasons cannot tell "very unlikely" from "eliminated", and a model that holds team strength fixed is too sure of itself near the ends.
Inputs, dated
- Loading…
Machines: read /data/nfl-playoff-leverage.json — every number on this page, with its method, inputs and limits. Its if_home_win and if_away_win arrays carry all 32 teams for every game, which is what the root-for guide reads.