MLB Playoff Predictor

How the MLB playoff format actually works

Twelve clubs, six per league. The three division winners are seeded 1, 2 and 3 by record among themselves, and the three best remaining records in the league take the wild cards at 4, 5 and 6.

The rule that trips people up: a division title always outranks a better record. The 2025 American League is the clean example. Cleveland won the Central at 88-74 and was the 3 seed. The Yankees finished 94-68, six games better, and were the 4 seed, because they finished second in the East. Six extra wins bought them nothing but a road trip.

Seeds 1 and 2 skip the first round entirely. The other four play a best-of-three Wild Card Series, 3 against 6 and 4 against 5, and the higher seed hosts all three games. There is no split. Then the bracket locks: the 1 seed plays whoever survives 4/5 and the 2 seed plays whoever survives 3/6. It does not reseed, so a 6 seed that pulls off an upset still gets the 2 seed rather than the 1. Division Series are best-of-five, and the Championship Series and World Series are best-of-seven.

How does a tie break work in MLB?

There is no tiebreaker game any more. MLB abolished Game 163 in 2022, the same year it expanded to a 12-team field, and ties have been settled on paper ever since. There were 12 one-game tiebreakers in the sport's history; the last was in 2021. The order:

1. Head-to-head record between the tied clubs
2. Record inside their own division
3. Record inside their own league
4. League record over the second half of the season
5. Working backwards through the schedule, game by game, until something separates them

Rule one does nearly all the work. Across the four seasons in the tool — 2022 through 2025 — every single tie that mattered was settled by head-to-head, and rule two has not been needed once. That is why the season series between two contenders in September is worth more than it looks: it is not just a game in the standings, it is the tiebreaker.

2025 is the season that shows why this matters. Toronto and the Yankees both finished 94-68. Toronto took the AL East, the 1 seed and a bye; the Yankees went to the Wild Card round — on an 8-5 season series. Detroit and Houston both finished 87-75 and Detroit got the last AL spot on a 4-2 series, which means Houston missed the playoffs entirely on a tiebreaker. Cincinnati took the final National League wild card over the Mets the same way, 4-2. Three ties, three head-to-head resolutions, one club eliminated by one. Any simulator that breaks ties with a coin flip gets that season wrong, and this one shows you the record it used.

Does it work? The backtest

A model you cannot check is a guess with a percentage sign on it. So here is the check, and you can rerun it yourself — scripts/backtest-mlb-playoffs.mjs in the source.

Seeding. Run the seeder over all four completed seasons of the 12-team era and it reproduces the real postseason field: all 12 clubs, in the right seed order, in both leagues, in every season. That includes all eleven ties it had to break, among them the three from 2025 above and the 2023 case where Miami and Arizona both finished 84-78 and Miami took the 5 seed on a 4-2 series.

Odds. Rewound to six dates in each season and scored against what actually happened, the model's Brier score is 0.111, against 0.240 for a baseline that just guesses the base rate at every club. From a 15 August cut it has 10 to 12 of the eventual 12 playoff clubs in its top 12; from mid-May, 7 to 10. That is the honest shape of it — in May this is a weak signal, and by September it is a good one.

Where it is wrong. When the model says a club is 70-90% to make it, that club actually made it about 64% of the time. It is overconfident in that band. The cause is not a mystery: the model has no idea about injuries, trade-deadline moves or September callups, so real seasons vary more than simulated ones do. We would rather print that than tune it away on four seasons of data and call it calibrated.

The model, in full

Most predictors call their model proprietary. Here is ours, in the order it runs:

pythag    = RS^1.83 / (RS^1.83 + RA^1.83)

strength  = (pythag x gamesPlayed + 0.500 x 60) / (gamesPlayed + 60)

p(game)   = log5(strengthHome, strengthAway), tilted to 0.534 at home

season    = play every remaining game as that weighted coin
            rebuild the standings, seed the field by the real rules
            run the bracket

odds      = repeat 10,000 times, count

Strength comes from run differential, not from won-lost record. A club's runs scored and allowed predict its future better than its own record does, because W-L folds in luck in one-run games and that luck does not carry over. This is the Pythagorean expectation, and 1.83 is the standard exponent for baseball.

Then it is regressed toward .500. In April a +30 run differential is mostly noise; in September it is mostly real. The model blends each club's Pythagorean rate with a .500 prior worth 60 games, so early-season results are taken with the salt they deserve. That constant is not a guess — it is the value that best predicts clubs' actual remaining records across 2022-2025, and the sweep is in the backtest script. Turning it off costs about 1.7 wins of accuracy per club per season.

Home-field advantage is measured, not assumed. Home clubs won 53.9% of the time in 2021, 53.3% in 2022, 52.1% in 2023 and 54.3% in 2025. The model uses 0.534, the average of those. It does not claim to model individual parks.

What it does not know: injuries, trades, callups, rest, travel, or who is starting tomorrow. A club that just lost its ace is, to this model, still the club its run differential says it is. Treat the number as the middle of a wide distribution rather than a forecast, and do not bet on it — you will not beat a market that prices all of the above.

Why every game is simulated instead of just the standings

A shortcut most simulators take is to give each club a win total and sort. That cannot work in baseball, because the tiebreakers need to know who beat whom. A simulated season that ends with two clubs at 94-68 has to be resolved by the head-to-head record in that simulation, not the real one. So each of the 10,000 runs plays out every remaining game, rebuilds the full standings from those results, and seeds the field the way MLB would. It is more work, and it is the only way the answer means anything in a season like 2025.

The same simulation runs the bracket, which is where the World Series column comes from. Best-of-three is close to a coin flip even for a strong favourite, which is why a 100-win club's title odds are far lower than its playoff odds, and why the bye is worth so much more than the seed number suggests.

Where the data comes from

Every game, date and score is from Retrosheet, whose licence permits commercial use outright and asks for one thing in return: the notice below, in its own words. Division alignment is entered by hand, because Retrosheet's team files carry league but not division.

Nothing here comes from MLB's Stats API or Baseball Savant. Both are excellent and both restrict themselves to individual, non-commercial use in terms we have read, which is why the tool stops at the most recent completed season instead of quietly using them. If that changes, this page changes with it.

Frequently asked questions

How does the playoff system work in MLB?

Twelve clubs reach the postseason, six from each league: the three division winners plus the three best records left over, who take the wild cards. The top two division winners in each league sit out the first round on a bye. Everyone else plays a best-of-three Wild Card Series, then it is best-of-five in the Division Series and best-of-seven from the Championship Series on.

How does the wild card work for MLB?

The three clubs in each league with the best records that did not win a division get the wild cards, seeded 4, 5 and 6. Division does not matter — all three can come from the same one. The 4 seed hosts the 5 seed and the 3 seed hosts the 6 seed, each in a best-of-three where the higher seed hosts all three games. There is no travel and no reward for being the road team.

How does a tie break work in MLB?

By arithmetic, not by a playoff game. MLB scrapped Game 163 in 2022 when it went to a 12-team field. Ties now break on head-to-head record first, then record inside the division, then record inside the league, then league record over the second half. The first rule settles almost everything: whoever won the season series takes the higher seed.

What is the 3 way tiebreaker in MLB?

When three clubs tie, MLB ranks them by their combined record against each other. If that separates one club, it takes the top spot and the remaining two go back to rule one and start again — head-to-head between just the two of them. That reset is the part most explanations skip, and it is why a three-way tie is not simply a sorted list.

If you want the same treatment for a trade argument rather than a playoff race, the fantasy baseball trade analyzer publishes its formula the same way. Both live in the tools hub.