The Giants led by four with two innings to bat. The Twins had three. That gap is the whole trade.

The position

The contract is “San Francisco wins by over 4.5 runs” on Kalshi’s Minnesota vs San Francisco spread ladder. It settles YES only if the Giants win by five or more, which is the same question as whether Minnesota covers +4.5, read from the other side of the ticket.

We entered in the top of the seventh with San Francisco ahead 5–1. That is a margin of exactly four against a 4.5 strike: one run from being wrong, zero runs from being right. NO was available at a 69.76¢ all-in average in Kalshi’s dollar mode, $2,000 across 2,866.58 contracts, $2,866.58 back if the margin held at four or shrank. Break-even 69.76%.

The asymmetry, counted in at-bats

Here is the whole reason for the trade, and it has nothing to do with either team being good.

San Francisco was the home team and San Francisco was ahead. A home team still leading after the top of the ninth does not bat in the bottom of the ninth. The game simply ends. That is not a tactic or a tendency; it is the rule.

So from the top of the seventh, the two clubs did not have equal opportunity to move the margin:

And the exception does not rescue the other side. San Francisco only gets a ninth if Minnesota has tied the game or taken the lead, at which point the margin is zero or negative and a contract needing a five-run Giants win is already dead. The bottom of the ninth is available to them exclusively in the states where they have already lost this bet.

Read as a race, it is simple: to settle YES, San Francisco had to outscore Minnesota by at least one more run, using one fewer inning to do it.

What one inning is worth in runs

An inning is not an abstraction. David W. Smith’s Retrosheet study of 122,906 games from 1901 to 2003 publishes scoring rates per inning, split by home and visitor. Those are the numbers that price this.

Minnesota’s three remaining innings are worth 0.473 + 0.470 + 0.450 = 1.393 runs. San Francisco’s two are worth 0.513 + 0.502 = 1.015 runs, and that is using the home rates, which are higher per inning. Scaled up to the 2026 run environment, which is about 2.25% livelier than the 1901–2003 average, the figures are roughly 1.42 and 1.04.

So the expected change in the margin from the top of the seventh was about minus four tenths of a run. Not “the margin probably holds.” The margin was expected to shrink. San Francisco being the better scoring side per inning was not enough to overcome having one fewer inning.

A note on honesty: the home ninth-inning rate of 0.415 is contaminated by exactly the effect this article is about, and it is not clear from the source whether skipped bottom-ninths are counted as zero-run innings or excluded. The comparison above never uses the home ninth, which is why it is the comparison worth making.

The ninth is the only inning home teams are bad at

Look at that table again, because it contains something odd. Across the full Retrosheet sample, the home team outscores the visitor in every single inning from the first through the eighth. Then, in the ninth, the sign flips: visitors 0.450, home 0.415.

A team that is better at scoring in all eight innings where the rules are symmetric becomes worse in the one inning where they are not. That is not a talent effect that appears at 10 PM. It is the structure. Smith attributes it directly: the game ends if and when the home team takes the lead, whereas the visitors keep batting no matter how many runs they score.

Two things truncate the home team’s scoring, and both cap a winning margin:

• The skipped ninth. Leading after the top of the ninth, they never bat. An entire inning of potential margin is deleted.
• The walk-off. Winning in their last at-bat, the game stops the instant they take the lead. They cannot run it up even in principle.

This is documented, not something we invented. SABR’s Home-Field Advantage states it plainly, the home team stops scoring when it takes the lead late, and frequently does not bat in the ninth at all, and the author excluded ninth-inning data from parts of that analysis specifically because the distortion would contaminate inning-by-inning comparisons. When a researcher throws out a column of data to protect their result, that column is telling you something. Here, it is the whole trade.

What nobody actually knows

We went looking for three numbers to nail this down and could not find two of them published anywhere. Saying so is part of the work.

How often does a home team skip the ninth? Not published, as far as we can find. What exists is a bound: home teams win about 54% of games, 54.2% across 1901–2011 in Pankin’s Retrosheet database, typically 53–55% by era. That is a hard ceiling, because a home team can only skip the ninth if it wins, and some share of home wins are walk-offs where it did bat. The true figure is somewhere below 54%, plausibly in the mid-to-high forties. That is a bound, not a measurement, and we are not going to print it as one.

Given a four-run lead after six and a half innings, how often does the final margin reach five? Also not published. Nobody conditions the margin distribution on game state. The unconditional distribution is known, roughly 29% of games end by one run, 18% by two, 15% by three, 11% by four, 8–9% by five, but that is a different question and we will not dress it up as this one.

How big is the truncation effect in margin terms? The mechanism is documented. The magnitude, expressed as “home winning margins are X runs narrower,” has not been quantified by anyone we could find. The arithmetic in the section above is our extension of documented work, not a citation of it. Treat it accordingly.

One more trap worth naming. Tangotiger’s win-probability table puts a four-run lead entering the top of the seventh at about .948 to win the game. That number is true and it is irrelevant. This was never a bet on who wins. It was a bet on by how much, and substituting one for the other is the most common way people lose money on run lines.

Why this matchup amplified it

The structural argument holds for any home team leading by four in the seventh. This particular pair made it better.

The side that needed to pull away had the worse offense of the two. San Francisco scored 4.21 runs a game this season; Minnesota scored 4.56. The team with fewer innings left was also the team less likely to use them.

And the Giants’ late innings are where their season has gone wrong. Their bullpen carried a 4.33 ERA, 22nd of 30, having converted 24 of 44 save opportunities, twenty blown saves. As of the day of this game they had no designated closer, with ninth-inning work distributed by matchup and availability. A bullpen that cannot protect leads is not a bullpen that extends them.

The bullpen ERA and save figures are dated Aug 24, 2026, roughly a month before this game, the most recent published version we could verify. We also note, as our own arithmetic rather than a published finding, that San Francisco’s runs scored and allowed imply a Pythagorean record near 70–86 against an actual 64–92: a six-win underperformance, the signature of a team losing close and late games.

What would have killed it

One swing. The margin was four, the strike 4.5, and a single San Francisco run in either of their two remaining innings puts it at five and flips the contract. A two-run homer does it with room to spare. There was no cushion whatsoever, which is precisely why NO cost 69.76¢ and not 90¢. The market was charging about thirty cents for two innings of Giants at-bats, and that is not obviously wrong.

The second risk is the one that catches people on run lines: Minnesota scoring does not help unless San Francisco stops. Giants add two, Twins add one, margin goes to five, contract settles YES even though the Twins “scored.” The trade is about the difference, never about either team’s runs alone.

Max loss was fixed at the $2,000 that went in. No stop, with two innings left there is nothing a stop can meaningfully protect, so the risk decision was the size at entry, and the position was held to settlement.

What happened

San Francisco scored two in the first, two in the second, one in the fifth. After that, nothing.

The two at-bats this trade counted on, the bottom of the seventh and the bottom of the eighth, produced no hits and one hit respectively, and zero runs between them. Minnesota added a run in the eighth, pulling the margin from four to three and moving it further from the strike. The Twins went down in order in the ninth.

And then the dash. San Francisco never batted in the ninth. Leading 5–2 after the top half, they did not come to the plate, and the game ended on a margin of three. The final at-bat that might have pushed the margin to five was not lost or wasted. Under the rules it did not exist.

That is the entire thesis, and it is also the literal reason this game finished where it did. The structure did not merely favour the position. It closed it.

Updates

• 9:22 PM PT · settled, Kalshi finalised the market: result no, expiration value 3, the YES side paid $0.0000. The NO position pays the full dollar. $2,000 in, $2,866.58 out, net +$866.58 (+43.3%), with no settlement fee. Nothing beat holding, NO peaked at a 99¢ bid and settlement paid 100¢.
• 9:31 PM PT, Minnesota went down in order in the ninth. San Francisco, leading 5–2, did not bat in the bottom half: the game ended on a margin of three, inside the 4.5 line. The final at-bat that could have pushed the margin to five did not exist under the rules. Grading follows once Kalshi settles the contract.
• 9:08 PM PT, Minnesota 2, San Francisco 5. The Twins have scored since we entered, so the margin is three, a full run further from the 4.5 line than when the position went on. NO is bid 99¢ against the 69.76¢ entry, with YES showing no bid at all. Nothing has been sold: the position is held to settlement, where a contract pays the full dollar and Kalshi charges no fee. Marked at 99¢ the $2,000 is worth about $2,838; settled it is $2,866.58.
• 8:45 PM PT, Entered. Minnesota 1, San Francisco 5, top of the 7th. NO at a 69.76¢ all-in average, $2,000 across 2,866.58 contracts.

Graded on the final margin, as written before the result was known. The entry, the reasoning and the risk were all published while the game was live and the outcome was not.