A hot start and a good team are not the same claim. A team can open 5–1 by consistently outplaying its opponents, or it can open 5–1 by winning the coin flips inside its own games — recovering the fumbles that bounce its way, forcing interceptions off tipped balls, and closing out one-score games at a rate the underlying scoring margin does not support. Both teams show a 5–1 record. Only one of them is likely to keep it.
Luck regression is the name for the second pattern. It is not a claim that the team is bad, and it is not an accusation that the wins do not count — every win on the schedule counts exactly once. It is a claim about which parts of a record are durable and which parts are the least repeatable inputs in the sport, dressed up as skill because the scoreboard does not separate them.
The gap between record and process
Win–loss record is the final output of a season, but it is built from inputs that vary enormously in how much they persist. Points scored and points allowed, adjusted for opponent, tend to be sticky: a team that is outscoring opponents by six points a game in September is usually still doing something similar in November. Turnover margin and one-score-game results are the opposite — they swing hard on small samples and pull back toward the field average as more games are played.
A luck-regression claim reads the box score for the gap between those two families of input. When the record and the process disagree — a good record built on shaky process, or a poor record built on solid process with bad puck luck — the claim is that the record moves toward the process, not the other way around, as the sample grows.
Turnover margin: the most mean-reverting stat in football
Turnover margin is repeatedly cited as the single most volatile team-level number that still correlates strongly with winning in the season it happens. A team recovering an outsized share of the fumbles on the field, or watching its opponents throw an outsized share of tipped-ball interceptions, is not usually repeating a skill — fumble-recovery rate in particular has long been treated as close to a coin flip at the team level once possession changes hands. The offense that has turned the ball over rarely so far this season is not necessarily protecting it better than the field; it may simply not have had the bounces go against it yet.
This is why a turnover-margin-heavy record draws the most attention from a luck-regression read. A team can be genuinely disciplined with the football and still be running well ahead of a repeatable rate, and the two explanations look identical in the standings.
Close-game record: the second leak
The second input a luck-regression claim checks is how a team has done in games decided by one score. Close-game outcomes are influenced by real factors — late-game execution, special teams, coaching in two-minute situations — but they also carry a meaningfully larger random component than blowouts do, because a handful of plays at the margin decide the final score. A team that has won nearly every one-score game on its schedule has usually benefited from more than a repeatable late-game edge; it has also needed the ball to bounce its way in exactly the moments that matter most for the final digit.
A team with a strong overall scoring margin that has also gone unusually well in one-score games is compounding two favorable draws into one record. Regression does not say the next close game is a loss. It says the historical base rate for sustaining a perfect or near-perfect one-score record is low, and the record should be expected to average back toward the team’s underlying scoring margin over the following games.
The Pythagorean win gap
Pythagorean win expectation — borrowed from Bill James’s baseball work and adapted for football — converts points scored and points allowed into an expected win total, independent of how the actual wins were distributed across games. Comparing a team’s actual wins to that expected total produces a gap. A team running several games ahead of its Pythagorean expectation is usually the same team piling up turnover luck, one-score wins, or both; a team running behind it is usually the mirror image, outscoring opponents overall while dropping a disproportionate share of the close ones.
The gap is a diagnostic, not a verdict. A team can sustain a positive gap for a full season, and single-season Pythagorean deviations are noisy on their own. What the gap does is point a reader toward the same two mechanisms above — turnover margin and close-game record — as the most likely explanation whenever a record and a scoring margin disagree by more than a game or two.
What a luck-regression lens would have to claim
Turning this mechanism into a lens — in the sense this series uses the word — means writing a claim that can be scored before the result is known: a defined turnover-margin or one-score-record threshold, a defined market (spread, moneyline, or total), a defined decision window, and a defined grading rule for wins, losses, pushes and no-calls. The claim would then need settled calls, an eligible sample, and a public gate — GRADING, PROVEN, FEED_ONLY or DECAYING, in the vocabulary this series uses elsewhere — before its record could be shown next to a number.
None of that scoring machinery is what this article is. This is the mechanism explainer: it says what luck regression is and why turnover margin and close-game record are the two levers to check. It is deliberately silent on any specific team, any specific week, and any specific win rate, because none of those figures are being asserted here.
What exists in the codebase today
The closest existing artifact to this claim is a model blueprint, not a graded lens: a Playbook (Py) entry keyed bp_turnover_regression_elasticnet, an ElasticNet regression trained on turnover-margin and turnover-luck features against the game spread. It is a modeling blueprint — a specification for a model that could be trained and evaluated — not a public claim with settled calls and a record. Its existence confirms that turnover-driven regression is a recognized modeling idea in this system; it does not confirm that any lens carrying that idea has a live, graded record today.
Sources and method
This article describes the luck-regression mechanism only: turnover margin as the most mean-reverting team stat in football, close-game record as the second source of unstable win accumulation, and the Pythagorean win gap as the diagnostic that flags when the two are compounding. It intentionally contains no live team result, no current lens win rate, and no fabricated record for any "luck regression" or "E-luck" lens. A record belongs on the live Lens Index once a lens with this exact claim and grading contract has eligible, settled calls to show — this explainer is not that record.
Expected bankroll growth at 55% edge
Expected geometric growth of a $100 bankroll under different Kelly multipliers across 1000 bets at p=0.55, decimal=2. Full Kelly maximises long-run growth but produces the deepest drawdowns; fractional Kelly trades growth for variance.
EV per $100 across win rate × odds grid
Expected value of a $100 stake at each combination of true win rate and market odds. Anywhere the cell is positive you have a long-run profitable bet; the magnitude shows how aggressive Kelly will size it.
