Here’s a scenario that plays out every baseball season. Two teams sit at 45-40. On paper, identical — both eight games over .500, both looking like solid contenders. You’d price a bet on them about the same.

But one of those teams has outscored its opponents by 90 runs on the season, and the other by 12. Same record. Not the same team. One is genuinely good; the other has been lucky, winning a pile of close one-run games that could just as easily have flipped. The market, if it’s pricing them by record, is about to hand you an edge on both.

The tool that separates the real team from the lucky one is Pythagorean expectation — one of the oldest and most reliable ideas in baseball analytics. If you bet baseball and you’re not using it, you’re reading the story instead of the truth.

What it is

Bill James, the founder of modern sabermetrics, stumbled onto this in the early 1980s editions of his self-published Baseball Abstract. He wasn’t working from a theory — he was staring at a spreadsheet and noticed that a team’s ratio of runs scored to runs allowed tracked its winning percentage far more tightly than its record tracked itself from one stretch to the next. After trying a few forms, he landed on squaring the numbers, which fit the real data almost eerily well. He named it “Pythagorean” because the sum-of-squares in the denominator reminded him of the theorem from middle-school geometry:

Win% = Runs Scored² / (Runs Scored² + Runs Allowed²)

Score more than you allow and the formula returns a winning percentage above .500. The bigger the run differential, the higher the expected win rate. Multiply by games played and you get expected wins — the record a team’s run differential says it earned.

The key mechanism: this is regression toward run differential, not toward .500. A team that wins ten straight one-run games has the same Pythagorean expectation as a team that loses ten straight, if their underlying runs scored and allowed are identical. Variance in close games is mostly noise. Run differential is mostly signal. The formula’s whole job is separating the two.

Why it actually works (not just “it fits”)

For two decades the formula was an empirical curiosity — it worked suspiciously well and nobody could say why squaring was right. That gap got filled in the 2000s, and it matters if you’re going to trust the thing with money.

The foundational answer came from mathematician Steven J. Miller in 2007. He showed that if you model each team’s runs scored and runs allowed as independent Weibull distributions — a standard tool in probability and reliability statistics — the Pythagorean formula falls directly out of the math as the probability that one team outscores the other. In plain terms: it isn’t an analogy to geometry at all. Under reasonable assumptions about how runs are distributed, it is the expected win probability. That’s the work most people are pointing at when they say the formula is “mathematically justified” rather than just observed. A later study extended the same logic to hockey, confirming it’s a general property of sports where chance plays a moderate role — which is also why basketball needs a much larger exponent (chance decides fewer basketball games) and football sits around 2.4.

The 1.83 refinement

James originally used an exponent of 2. Decades of testing found the formula predicts real wins slightly more accurately with a lower exponent, and the widely used modern standard is 1.83:

Win% = RS¹·⁸³ / (RS¹·⁸³ + RA¹·⁸³)

More advanced versions (”Pythagenpat”) calculate a dynamic exponent based on each team’s run-scoring environment, which matters in unusually high- or low-scoring contexts. But 1.83 is the sharp, simple standard, and it’s all you need to start. However you compute it, the accuracy is the striking part: a team’s Pythagorean record typically lands within 2 to 3 wins of its actual record over a full 162-game season — remarkable for a formula that ignores lineups, injuries, bullpen usage, and clutch hitting, and looks only at runs in and runs out.

Where the betting edge lives

Here’s the part that pays. When a team’s actual record diverges meaningfully from its Pythagorean record, that gap tends to close. The team regresses toward what its run differential says it is. And the market is slower than you’d think to price this, because most bettors pattern-match off the standings page rather than running run-differential numbers before they click.

  • A team outperforming its Pythagorean record — more wins than expected — has usually been winning close games at an unsustainable rate. Close-game luck doesn’t persist. This team is a cool-off candidate, and if the market still prices its shiny record, it’s overvalued → a fade.

  • A team underperforming — fewer wins than its runs deserve — has been losing close games or getting buried in a few blowouts that inflated runs allowed. This team is a heat-up candidate, and if the market prices its mediocre record, it’s undervalued → a back.

The bet isn’t “the math says they’re good, so back them.” It’s “the market is pricing the record, the record is misleading, and the correction is coming.” You’re trading the gap between perception (record) and reality (run differential), and betting reality wins.

The textbook case: the 2023 Padres

If you want to see the gap at its most extreme, look at the 2023 San Diego Padres — and the numbers are more dramatic than most people remember.

San Diego finished 82-80, a .506 record, and missed the playoffs by two games. But they scored 752 runs and allowed 648 — a +104 differential. Run that through the formula and you get a .574 Pythagorean win percentage, which is an expected record of 92-70. Baseball-Reference lists exactly that: Pythagorean W-L, 92-70.

That’s a ten-win gap between what the run data said the team was and what actually went in the standings. Excluding strike- and pandemic-shortened years, that ranks among the worst underperformances for a positive-run-differential team since 1901. The Padres had a roster stacked with stars and one of the best run-prevention units in the National League — and they went 6-22 in one-run games and 0-11 in extra innings. The public watched a “disappointing .500 team.” The math saw a 92-win team that couldn’t catch a break in close games.

For a bettor, that’s the whole lesson in one team: the standings said “mediocre,” the run differential said “top wild-card seed,” and the truth was the run differential.

The other direction: watch regression actually happen

The formula cuts both ways, and 2026 offered a clean live example of the overperforming side — and of the regression actually landing.

In late May 2026, roughly a third into the season, the Cincinnati Reds were one of the quiet names on every “the standings are lying” list — sitting several games above what their run differential supported, propped up by an above-water record in one-run games. The Pythagorean read was blunt: this isn’t sustainable, expect regression.

Fast-forward to midsummer and that’s exactly what happened. By early July the Reds’ actual record and their Pythagorean record had converged — both now reflecting a sub-.500 team (Baseball-Reference had them around a 37-49 Pythagorean mark, closely tracking their real record). The early-season overperformance didn’t hold. The close-game luck ran out, the record fell back to what the runs said all along, and anyone who faded the “20-something-win team with a negative differential” got paid as the correction arrived.

That’s the more useful version of the lesson, because you can watch the mechanism work end to end: the formula flagged an unsustainable gap in May, and by July the gap had closed in the predicted direction.

How to use it

Rank teams by their Pythagorean gap. Pull every team’s runs scored, runs allowed, and record; compute expected wins; sort by the size of the gap. Big positive gap (lucky, overvalued) → fade watch list. Big negative gap (unlucky, undervalued) → back watch list.

Only bet the gaps the market hasn’t priced. A team everyone knows is unlucky may already have adjusted lines. The edge is the team whose bad record is still driving a soft price. This is the same principle behind our scanner — the value isn’t in the data everyone has, it’s in the spots where the price hasn’t caught up to it yet.

Lean on it hardest for season win totals. The gap is a large-sample signal, so preseason and mid-season win-total markets are its natural home. A team that badly underperformed its run differential is a classic win-total over candidate as luck normalizes.

The honest limits

It flags the gap; it doesn’t explain it. Sometimes a team beats its Pythagorean record not by pure luck but with a genuinely elite bullpen that wins close games repeatably. The formula can’t tell those apart — you still have to look under the hood.

Blowouts distort run differential. A couple of 15-1 losses wreck a team’s runs-allowed number and make them look worse than they are, because the formula weighs lopsided games as heavily as close ones.

It needs a real sample. Don’t treat a five-game stretch as a signal. Most practitioners don’t trust it as a forward-looking read until a team is 30-plus games in, and even then it’s a probabilistic lean, not a lock. Recent testing pegs the first-third-of-season Pythagorean mark as correlating with final record at roughly 0.74 — a strong signal, with a margin of error around nine wins. A signal, not a crystal ball.

The market often already knows. Sharp bettors and books run this exact formula. On heavily-traded markets the gap may already be priced. The edge lives in softer spots — win totals, smaller-market teams, early-season prices before the sample fills in.

The takeaway

A win-loss record tells you what happened. Pythagorean expectation tells you what should have happened — and, more usefully, what’s likely next. When those two numbers disagree, the record is usually the one that’s misleading, inflated or deflated by close-game luck that doesn’t repeat.

Compute the gap. Find the teams furthest from their run differential. Cross-check against market prices for the ones the market hasn’t corrected yet. Then bet on reality catching up to perception — because over a baseball season, as the 2023 Padres and the 2026 Reds both show in opposite directions, it almost always does.

The record is the story everyone’s reading. The run differential is the truth. Overrated or undervalued — bet the difference.

Formula: Win% = RS^1.83 / (RS^1.83 + RA^1.83), with expected wins = Win% × games played; James’s original used exponent 2, and dynamic-exponent variants (Pythagenpat) are marginally more accurate. Statistical grounding via Steven J. Miller’s 2007 Weibull-distribution derivation. 2023 Padres figures (82-80 actual, 92-70 Pythagorean, 752 RS / 648 RA) per Baseball-Reference. 2026 Reds figures reflect a late-May overperformance flagged by Pythagorean trackers and a subsequent convergence to a sub-.500 Pythagorean record (~37-49) by early July per Baseball-Reference; current-season numbers move continuously — verify live. Pythagorean records typically fall within 2-3 wins of actual over a full season; first-third-season Pythagorean win% correlates with final record around 0.74 with ~9-win error. The formula flags over/underperformance but does not explain its cause; regression is a large-sample tendency, not a single-game guarantee; efficient markets may already price known gaps. Nothing here is betting advice.