The Assets, Not the Logos: Thirteen Years of NFL Win-Total Bias and the 2026 Reset

Overview

Across the last thirteen NFL seasons (2013–2025), the preseason win-total market has been systematically wrong about specific franchises in the same direction year after year. One franchise hit the under on its preseason win total 11 of 13 times. Three more hit it 10 of 13. On the other side, two franchises hit the over 10 of 13. A Monte Carlo simulation confirms that this clustering of extreme franchises is unlikely under random independence (p = 0.030). The bias is real and the bias is structural.

A natural-experiment case (the Patriots, examined across four distinct organizational eras) shows the biases are tied to specific organizational assets, primarily the head coach, rather than to franchise identity. The 2025 season produced three more confirming cases: when transformative new head coaches arrived at chronically under-performing franchises, the bias did not just dissolve, it reversed dramatically. The 2026 offseason then disrupted eight more of these franchise patterns simultaneously, producing what may be the cleanest forward test of an analytical thesis in modern NFL history.

This note lays out the evidence, states in advance what the framework expects each 2026 test to show, and commits to grading every expectation publicly in January. It contains research and expectations, not wagering advice.

Finding 1: The bias clusters

Sorting all 32 franchises by their over/under hit rate across 13 seasons reveals two distinct populations at the tails. Eleven franchises sit beyond the 60/40 mark in one direction, with the remaining twenty-one franchises clustering near a 50/50 split.

The implied mispricing is large. At standard -110 juice, breakeven is 52.4%. The extreme franchises clear or fall short of breakeven by 9–37 percentage points. The natural question is whether this kind of dispersion can occur by chance across 32 franchises rolling 13 times each. The next finding addresses that directly.

Finding 2: The clustering is statistically robust

Skeptics will reasonably ask whether the observed pattern is just luck. With 32 franchises rolling 13 seasons each, you would expect some franchises to land at extreme over rates by chance alone. The right test is not “is the most extreme franchise unusually extreme” but “are there unusually many extreme franchises clustered together.”

A Monte Carlo simulation answered this directly. We drew 100,000 simulated leagues, each with 32 franchises rolling 13 seasons of independent coin flips at the empirical league rates observed across our actual data (P(over)=0.457, P(under)=0.507, P(push)=0.036). For each simulated league we counted how many franchises landed at 10 or more in the same direction.

Three things to note. First, the individual extreme franchises (NYG at 2-11, KC at 10-3) are not unusual on their own; with 32 franchises and 13 seasons each, the most extreme franchise should be about that extreme by chance. Second, what is unusual is the count of simultaneously extreme franchises: the data shows 6 franchises at 10+ in the same direction where chance predicts about 2.4, a 3.0% probability under independence. Third, this clustering is exactly what an asset-driven hypothesis would predict: if certain organizational situations produce persistent multi-season biases, you would see groups of similarly-situated franchises deviating together, not isolated random outliers.

The Monte Carlo rejects the null of “independent coin flips” at conventional significance. It does not by itself prove that organizational state is the mechanism. For that, we turn to the cleanest natural experiment in the dataset.

Finding 3: The bias is the asset, not the franchise

The Patriots provide the cleanest natural experiment in the dataset. New England has produced four distinct organizational eras in this window, each with a different bias signature:

• Brady years (2013–2019): 5-0-1 over. The market literally could not price the team high enough; even the 12.5 total in 2017 was cleared.
• Post-Brady Belichick (2020–2023): 1-3 under. The bias did not gradually decay, it flipped immediately and decisively the year Brady left.
• Mayo year (2024): 1 under. Belichick gone, no replacement asset yet, franchise drifted further down.
• Vrabel year (2025): 1 over, by 8.5 wins. New England finished 14-3 against a 5.5-win preseason total. The largest single-season variance in the entire dataset.

The implication is structural. The market anchors preseason win totals on prior-year performance but underweights how much that performance depended on specific personnel. When personnel changes, bias dies. When the right personnel arrives, bias can reverse violently. KC overs are a Mahomes–Reid pattern. PIT overs were a Tomlin pattern. The unders franchises share a different pattern: chronic organizational instability (coaching changes, QB carousels, mid-season firings) that the market keeps trying to price optimism out of, year after year.

Finding 4: 2025 confirmed the framework in real time

The Patriots case is one historical natural experiment. The 2025 season produced three more, all at once. Going into 2025, three franchises with deeply entrenched under-bias histories hired transformative new head coaches. The asset-continuity hypothesis predicts that the bias should dissolve, and possibly reverse, with the new asset in place. All three hit the over, by margins that placed them among the biggest outliers in the entire dataset:

Three for three. The framework predicted these biases should reverse when the right organizational asset arrived. They did, in the same season, with magnitudes that would individually have been remarkable. This is the first prospectively-testable evidence that the bias-from-organizational-assets hypothesis produces predictive results, not just retrospective curve-fits.

Finding 5: The biases are mostly strengthening

Splitting each extreme-bias franchise into an early era (2013–2018) and a late era (2019–2025) reveals that the biases have mostly intensified over time rather than fading toward market efficiency. The dominant exception is Kansas City, where the market has clearly caught up to the Mahomes narrative.

Two structural observations emerge. First, the market appears to adjust aggressively when a franchise becomes a cultural narrative (KC) but conservatively when a franchise wins quietly (PIT, BUF, SEA). The price-setting process is biased toward fame more than toward consistency. Second, several unders franchises (NYJ, ATL) actually accelerated their underperformance in the late era. The 2019–2025 window contained the most extreme franchise mispricings of the entire dataset, despite a decade of data being available to correct them.

Finding 6: 2026 disrupted eight of the eleven

Ten head coaching changes were announced this offseason, tied for the most ever in a single NFL offseason. Of the eleven franchises in the extreme-bias clusters, eight have a new head coach, a new starting quarterback, or both heading into 2026. Each represents a live test of whether the historical bias was franchise-level (and should persist) or asset-level (and should dissolve or reverse, consistent with the 2025 evidence).

Figure 5. Forward-watch table, verified as of July 14, 2026. NEW = appointed for 2026; Y2 = second season under current appointee; (continuous) = no change. LAR has 10 completed seasons in the dataset (2013–2015 St. Louis era partial).

The 2026 Test Board

With the framework laid out and the 2025 results as fresh evidence, here is what the framework expects each live test to show in 2026. These are falsifiable expectations, stated in advance, that we will grade publicly in January. They are not wagering recommendations; where your own read differs, that divergence is exactly the thing worth investigating.

A note on the in-sample question

A sharp reader will ask: if these franchises were identified by looking at thirteen years of historical data, of course the historical patterns would show a high hit rate; that is in-sample selection by construction. The fair test is out-of-sample, and the 2025 season provided exactly that. For each extreme-bias franchise, the framework made a directional prediction before the season based on its asset-continuity refinement (continuity = bias persists; transformative new asset = bias may break or reverse). The results:

• KC: predicted OVER (Mahomes-Reid continuity). Actual: UNDER (Mahomes ACL injury). MISS.
• PIT: predicted OVER (Tomlin continuity). Actual: OVER. HIT.
• BUF: predicted OVER (Allen-McDermott continuity). Actual: OVER. HIT.
• SEA: predicted OVER (post-Carroll bias survived 2024). Actual: OVER, won Super Bowl LX. HIT.
• CHI: would have predicted OVER given the Ben Johnson hire (asset reset). Actual: OVER. HIT.
• JAX: would have predicted OVER given the Liam Coen hire (asset reset). Actual: OVER. HIT.
• NYG: predicted UNDER (no transformative reset). Actual: UNDER (Daboll fired mid-season). HIT.

Six hits, one miss, with the miss attributable to an unforeseeable mid-season injury rather than a framework failure. The remaining test cases (PIT post-Tomlin, BUF post-McDermott, NYG post-Daboll, ATL post-Morris) are the live 2026 tests graded above.

Caveats and limitations

Thirteen seasons is meaningful but not infinite. The Monte Carlo result (p = 0.030) is strong enough to reject the null of random independence at conventional thresholds but does not rule out other null hypotheses, such as “franchises have stable but random multi-season biases unrelated to organizational continuity.” The asset-continuity hypothesis is consistent with the data and supported by the Patriots case plus the 2025 three-for-three, but neither piece of evidence rises to formal causal proof on its own.

Expectations involving KC’s post-ACL Mahomes are particularly speculative. Recovery from major knee surgery is highly individual, and even a healthy return does not guarantee the market’s adjusted line will be wrong in either direction. The KC case is included for completeness as the most uncertain entry on the board.

Throughout this note we refer to “the market” rather than “Vegas” specifically. Modern win-total lines are shaped by initial bookmaker opinion plus the aggregate of betting action and limit-driven adjustments. The mispricing observed here is a property of the resulting market price, not necessarily of any single sportsbook’s opinion.

Methodology note

The AGINT Intelligence Graph is a multi-source NFL data product covering 13 seasons (2013–2025), spanning game-level outcomes, multi-book line snapshots, preseason win totals for all 32 teams, weather, advanced statistics, and intelligence assessments from a collection-agent layer. Regular-season records used in this analysis were computed directly from game-level outcomes in the graph, with the same playoff-exclusion logic applied uniformly across all 13 seasons. Monte Carlo simulations were run with 100,000 iterations, drawing each simulated team-season from the empirical league-rate distribution; the figure above was regenerated from a fresh 100,000-iteration run at publication, reproducing the original result. The graph preserves source provenance and temporal validity for every fact; this analysis is one of a range it supports.

Takeaways

• The preseason win-total market carries persistent franchise-level mispricings. Across 13 seasons, six franchises hit 10+ of 13 in the same direction. Monte Carlo gives p = 0.030 against random independence.
• The bias is structural, not random. Four Patriots eras plus three confirming 2025 cases (CHI/JAX/NE) demonstrate that bias dies when the asset that produced it departs and can reverse violently when the right asset arrives. The assets, not the logos.
• 2025 was the first true out-of-sample test. The framework correctly anticipated directional outcomes in 6 of 7 identifiable cases, including that the Bears’, Jaguars’, and Patriots’ historical patterns would break under their new head coaches.
• 2026 is the forward test, and it will be graded. Eight of the eleven extreme franchises enter the season disrupted. The franchises whose key assets remain (LAR, SEA) are the cleanest persistence tests; the ones that just lost their anchor (PIT, BUF, NYG, ATL) are the cleanest reversal tests; the ones whose bias already broke (CHI, JAX) are the cleanest new-normal tests. Judge 2026 by the assets that arrived, not the franchise records they joined.