The Late Innings Constant Model: The Gap Between A First Five Line And A Full Game Line
A sportsbook posts two totals on the same baseball game. One settles after five innings and one settles at the end. The difference between them is a statement about the last four innings, and it is a statement that can be checked against every game played this season.
The Late Innings Constant Model checks it. There is one question: how many runs arrive after the fifth inning, and what does that number depend on?
Oriole Park at Camden Yards, Baltimore. Late innings are where a full game total separates from a first five total, and this season the size of that separation has been remarkably stable. Photo: Quintin Soloviev, CC BY 4.0, via Wikimedia Commons.
The Baseline
Every 2026 regular season game with a Final status and at least five innings on the linescore, from March 1 through September 3. 2,108 games. For each one, the combined runs through five and the combined runs from the sixth inning onward.
| Segment | Mean combined runs | Share of the game |
|---|---|---|
| Innings 1 through 5 | 4.991 | 55.68% |
| Innings 6 onward | 3.972 | 44.32% |
| Full game | 8.963 | 100% |
Restrict to the 1,923 games that finished in exactly nine innings and the split is 5.105 and 3.780, a 42.54 percent late share. Extra innings account for the difference and they are what pull the unrestricted figure up to 3.972.
The Finding: It Does Not Move
The interesting part is not the average. It is what happens when the first five innings are conditioned on. Sort all 2,108 games by what the combined score was through five and look only at what happened afterwards.
| Combined runs through 5 | Games | Mean runs after the 5th | Mean final total | Final under 7.5 |
|---|---|---|---|---|
| 0 or 1 | 270 | 4.03 | 4.65 | 83.7% |
| 2 or 3 | 524 | 3.82 | 6.37 | 70.4% |
| 4 | 261 | 3.90 | 7.90 | 55.9% |
| 5 or 6 | 438 | 4.09 | 9.55 | 33.3% |
| 7 or more | 615 | 4.02 | 13.10 | 3.4% |
Read the third column top to bottom. 4.03, 3.82, 3.90, 4.09, 4.02. The combined score through five ranges from under one run to over seven across these buckets, a spread of more than twelve runs in the final column, and the number of runs scored after the fifth inning does not respond to it at all. The largest gap between any two cells is 0.27 runs and there is no direction to it.
A 1-0 game through five and a 9-4 game through five produce the same last four innings. Whatever a starting pitcher did, whatever the weather was doing, whatever the two lineups looked like early, once the bullpens are in the run rate reverts to the same place.
What The Model Says A Line Gap Should Be
If the late segment is a constant then the model has one rule, and it is a subtraction.
full game expectation = first five expectation + 3.97
The constant is not identical for every club. Sorting all 30 teams by the mean combined runs after the fifth in their own games gives a range from 3.50 to 4.74, which is a real spread and a much smaller one than most people would guess.
| Lowest late innings totals | Runs after the 5th | Highest late innings totals | Runs after the 5th |
|---|---|---|---|
| Texas Rangers | 3.50 | Washington Nationals | 4.74 |
| Boston Red Sox | 3.61 | Athletics | 4.55 |
| Houston Astros | 3.61 | Chicago Cubs | 4.41 |
| Los Angeles Angels | 3.65 | Colorado Rockies | 4.36 |
| Cleveland Guardians | 3.66 | Pittsburgh Pirates | 4.26 |
Tonight's Board Through The Model
Five games on the September 4 slate carry both a posted first five total and a posted full game total. The model's only interest is the difference between them.
| Game | First five line | Full game line | Implied late innings | Model late innings | Difference |
|---|---|---|---|---|---|
| Yankees at Padres | 4.5 | 7.5 | 3.00 | 3.80 | -0.80 |
| Red Sox at Orioles | 4.5 | 8.0 | 3.50 | 3.86 | -0.36 |
| Athletics at Mariners | 3.5 | 7.5 | 4.00 | 4.22 | -0.22 |
| Tigers at Guardians | 4.5 | 8.5 | 4.00 | 3.86 | +0.14 |
| Rays at Rangers | 4.5 | 8.5 | 4.00 | 3.65 | +0.35 |
Four of the five sit within four tenths of a run of the model, which is close enough that the model is not saying anything about them. The exception is the first row. Yankees at Padres carries a three run gap between its two totals, and the season figure for those two clubs is 3.80 combined runs after the fifth, New York at 3.93 and San Diego at 3.67.
Read plainly, that means the market is pricing the last four innings of this specific game about eight tenths of a run below what either club has produced all season. Two good bullpens is the obvious explanation, New York at a 3.24 team ERA and San Diego at 3.94, and the model has no way to see a bullpen. What the model can say is that this is the widest disagreement on the board and that the disagreement points at the full game under being the expensive side rather than the cheap one.
Where The Model Stops
It stops at three places and they all matter.
A line is not a mean. A posted total sits near the median of a right skewed distribution, and comparing it to a mean overstates the gap by a few tenths in exactly the direction seen here. Some of the 0.80 on the Yankees row is that and not a market opinion.
The constant is a season figure applied to a specific night. A team's late innings number is built from every reliever it has used all year, including ones who are no longer on the roster and ones who threw thirty pitches last night. The model has no bullpen availability input at all.
Extra innings are counted at face value. The automatic runner rule inflates the late segment in the 185 games that went past nine, which is why the unrestricted constant is 3.972 and the nine inning constant is 3.780. Which of those is the right number to compare a line against depends on how the book settles extras, and the model does not adjust for it.
The finding that survives all three is the one in the second table. Runs after the fifth inning do not care what happened before it. A model of the first five innings is a model of the first five innings, and it says almost nothing about how the game ends.
Data note: Results come from the MLB Stats API for the 2026 regular season, every game with a Final status and at least five innings on the linescore from March 1 through September 3, 2026, deduplicated by gamePk to 2,108 games. First five runs are the sum of the home and away runs in linescore innings one through five as published. Late innings runs are the final combined score minus that figure, extra innings included at face value. Team late innings figures are the mean of the combined late runs across every game that team appeared in, home and away, and are in sample. Posted lines are the public first five and full game totals on the September 4, 2026 board at 2:00 pm Eastern, taken from a single book. Implied late innings is the full game line minus the first five line. Model late innings is the unweighted mean of the two clubs season figures. No park, weather, bullpen availability, starting pitcher or lineup adjustment is applied. No result of this model is a recommendation to bet anything.