Playoff odds

How this board works

The numbers come from a 2,000-run simulation of the games that have not been played yet. Team strength is ESPN's Football Power Index, which is how many points a team is expected to win by against an average opponent on a neutral field. The page asks ESPN for that table, the CFP committee top 25, and the scoreboard. If the committee ranking is missing, the AP top 25 is used instead. If a feed fails, the page shows the last snapshot that saved, with the time it was saved.

Each remaining game, including a conference title game, gets a win probability before it is simulated. The probability starts from the FPI gap on a normal curve with a 14-point standard deviation. The home team gets 2.5 points unless ESPN lists the site as neutral. When ESPN posts a point spread, or a moneyline when there is no spread, 50 percent of the probability comes from that line and the rest from FPI. The posted line already includes home field, so the blend does not add home field a second time. When a game is in progress, the line and the FPI gap are set aside and the simulation uses ESPN's live win probability on the last play. If that probability is missing, the page estimates it from the score, the clock, and the pregame blend. After a game ends, the result is in the record and the next refresh simulates only what is left.

Conference games update conference records. A championship ESPN has already paired is played as listed, including a game where one team is already in and the opponent is still TBD. If both spots are open, the two teams with the best conference records in that simulation play it. Wins come first, then fewer losses. When the records are the same, each team is equally likely to get the spot, so a higher power rating does not take the game every time. The Sun Belt is the one ESPN still splits into East and West, so each division sends its own leader. On a home-site championship, the better record hosts. The winner is the conference champion.

The playoff is the 12-team field: the five conference champions with the best résumé get the automatic bids, and the next seven teams fill the at-large spots. The four best conference champions get the first-round bye. An independent can take an at-large spot. A bye only goes to a conference champion, so an independent's conference and bye columns are 0%.

The résumé is the record plus who was played. A win is worth 8 points and a loss costs 12. Each point of the opponent's FPI on a win adds 0.5 points, so a win over a strong team counts more than a win over a weak one. A loss to a strong team keeps 45% of that opponent's FPI, which is why losing to a good team is not the same as losing to a bad one. The team's own FPI adds 1.15 points per point of rating. A team in the current top 25 also gets 0.85 points for each spot above No. 26, a head start the results can overcome. Each simulation adds a random shake, typically about 6 points. That résumé is the committee's ballot for that season. It is not a prediction of any voter's actual ballot. Before any line is blended in, the same résumé is what lines this board up with ESPN's published playoff, bye, and conference odds. Where a number here differs from FPI, it is because the betting line and the live score moved the game probabilities.

If they win out uses the same simulated seasons and the same committee shakes. The only change is that this one team wins every remaining game, and wins the conference title game when those wins put it in that game. The column is how often the team is in the 12-team field in those seasons.

National-title odds play the bracket from those seeds. Seeds 5 through 12 play at the higher seed. The quarterfinals pair 1 with the winner of 8 vs. 9, 4 with the winner of 5 vs. 12, 2 with the winner of 7 vs. 10, and 3 with the winner of 6 vs. 11. Semifinals and the final follow. Those rounds are on a neutral field and use the same win curve.

The board is cached for about 90 seconds, so a refresh during a game picks up the new live probability without rebuilding on every visit. The same random draws are reused every time, so the odds stay put until a score, a spread, or the ranking changes. A scheduled job saves a snapshot for the arrows. Records are completed games on the scoreboard, so they can be ahead of the ranking published earlier in the week. With 2,000 runs, a probability near 50% is steady to about one percentage point.