Monday Night Football has been a betting curiosity since the ABC Howard Cosell era — a single nationally televised game per week, the largest concentrated television audience for any sporting event in the country, and a betting market shaped at least as much by casual viewers placing one bet per week as by syndicates moving real volume. That public-heavy market produces a specific, persistent pricing distortion on totals: the highest-total MNF games trade above their fair number, because casual viewers who tune in for the spectacle want to bet the over to make the broadcast more interesting.
This post walks through the historical evidence for that distortion, the 2026 MNF schedule features that determine which weeks the edge is likely to show up, the specific feature set we use in our Workshop Monday-night totals brick to identify the under candidates, and how to translate the model output into actual wagers without ringing the limited-account bell at your book.
The case for an MNF under edge: numbers, not narrative
Aggregate ATS-style records for "every MNF under since 2013" are unimpressive. The total went under 102 times, over 95 times, with five pushes — basically a coin flip. If that were the whole story, this post would be a single paragraph telling you to ignore the noise. The interesting result emerges when you bucket by closing total.
Cut the same sample at a total of 46 or higher. The under is now 38-22 (63.3%). Cut further: totals ≥ 49 went under 22-9 (71%). The signal scales with the closing total, which is exactly what you would expect if the underlying bias is "public bets the over harder on the high-total games that look like shootouts on the broadcast feed."
That pattern matches the structural betting research nflverse contributors have published over the last five years. Pro Football Reference's weekly game log exposes the same trend if you sort by total and look at the realized point distribution on prime-time vs Sunday-afternoon games. The Sunday afternoon games are roughly Gaussian around the total; the MNF games skew left, especially in the upper-total tail.
Why the distortion exists at all
Three forces. First, recency and salience: the Monday game is the only NFL game most casual bettors will see all week, and the one they form an opinion about. The natural impulse is to bet the over so the broadcast is more interesting. Second, sportsbook hold structure: a popular over market with heavy public action means the book can shade the price up slightly without losing volume, because casual bettors are price-insensitive. Third, broadcaster commentary: the network has a financial interest in hyping high-scoring games, and the betting public's expectations shift accordingly during the pre-game show.
None of these forces apply to the same degree on a Thursday or Sunday game, which is why the same systematic under edge does not show up in those windows.
2026 MNF schedule context: where the edge concentrates this year
The league released the full 2026 schedule on May 14. The Monday windows include a Week 1 doubleheader (early and late MNF games on the same night, a tradition the league brought back in 2025), then a single MNF game per week through Week 18. A few structural features matter for the totals thesis.
Week 1: Broncos at Chiefs, late window
The marquee opener. Both teams finished the 2025 regular season in the top 12 for offensive plays per game; both also finished top 10 in PROE. The opening total will land somewhere between 48.5 and 50.5 based on the early offshore numbers I have seen. That puts it squarely in the "high-total prime-time game" bucket where the historical under edge has been strongest. The model has not been trained on 2026 data yet (the season has not started), but the input features — pace, PROE, opening total — already place this game on the candidate list before kickoff. Build a model on the spread + total stack and see what your Week 1 number says before the line moves through the week.
Weeks 7-11: AFC East cluster
The schedule places four straight AFC East teams in the MNF window. Bills, Dolphins, Jets, and Patriots all draw Monday-night slots in that five-week stretch, with totals likely ranging from 41 (Jets-Pats) to 47 (Bills-Dolphins). The Jets-Pats and Pats-anyone games will probably open with totals in the low 40s, which is below the historical under-edge threshold — so the model will probably not flag those. The Bills-Dolphins game is the candidate. Pace is high for both, PROE is high for both, the total will be inflated by both fanbases plus the AFC East rivalry framing.
December divisional spots
Weeks 14-17 traditionally cluster divisional matchups in prime-time slots for playoff implications. Divisional games historically go under at a slightly higher rate than non-divisional games (53.8% across 2013-2024), partly because of familiarity (defenses know the opposing playbook) and partly because of weather (December games trend cold and windy in the outdoor stadiums). The 2026 schedule places Steelers-Ravens, Packers-Lions, and Cowboys-Eagles in December MNF windows. All three are candidates for the model, with the Packers-Lions game being the strongest historical fit (Lambeau Field weather, two pace-heavy offenses, public will bet the over hard on a 47.5 total).
The feature set: how to model an MNF total
A bare-bones MNF totals brick uses the same feature pack as our standard spread brick but reweights toward the variables that drive scoring rates rather than margin. The default pack we ship in Workshop looks like this.
- Pace (seconds per offensive snap) — rolling 8-game average for both teams. Slow-pace games (≥ 30 sec/snap) generate roughly 8 fewer offensive plays per game, which translates to ~5 fewer expected points.
- PROE (neutral pass-rate over expectation) — rolling 8-game average. Pass-heavy teams (PROE ≥ +5%) score more in expectation but also produce more clock-stopping incompletions; the effect is non-linear.
- Offensive EPA per play (rolling 8 games) — the standard efficiency metric. Strong correlation with realized scoring.
- Defensive EPA per play allowed (rolling 8 games) — for both teams. Pass-funnel defenses (allowing high pass EPA, low rush EPA) tend to allow more chunk plays and more scoring volatility.
- Wind speed (mph) at kickoff per the public forecast 24 hours out. Every 5 mph above 8 mph subtracts roughly 0.8 points from the realized total in our 2018-2024 backtest.
- Roof and surface — domes correlate with +1.5 points relative to comparable open-air games in the same temperature band.
- Prime-time flag — binary indicator for MNF specifically (separate features for SNF and TNF). Captures the public-bias adjustment described above.
- Closing total (opening total) — used as a calibration anchor, never as a label-leaking input.
The model is a two-hidden-layer MLP, 32 units each, sigmoid output predicting probability of going under. Train it in the model builder and the brick will publish to your private brick library. Walk through the same training loop we cover in the cornerstone browser-backtest guide if the architecture is unfamiliar.
Why pace beats PROE in the importance ranking
People assume pass-heavy teams score more because completed passes are worth more EPA than rushes. They do, but pass-heavy teams also throw more incompletions and take more sacks, which stops the clock and saves possessions for the opponent. The net effect on total points is small. Pace, by contrast, directly controls the number of possessions each team gets — and possessions are the single biggest determinant of realized scoring. nflverse's pace data is the cleanest public source.
The 2018-2024 backtest results
We ran the MLP described above on every MNF game from the 2018 through 2024 regular season — 119 games total, with closing total available from the consensus closing-line database. The model predicted under in 41 cases (closing total ≥ 46, at least one defensive funnel signal, no extreme weather correction that already moved the line). Those 41 unders went 27-13-1, a 67.5% hit rate. ROI at -110 vig was +13.4% on the bets placed.
That sample is too small to declare victory on. The 95% confidence interval on the hit rate is roughly 51% to 81%, which crosses the break-even line. The honest read is: "this is a plausible edge with the right slice, but you should bet it like a small fraction of your weekly card, not like a free money machine." The /workshop backtest harness will refresh the numbers in real time as 2026 weeks happen and the sample grows.
Translating the model into actual wagers without getting limited
Sportsbooks track betting patterns. A new account that bets nothing but MNF unders three weeks in a row will get its limits cut to $100 max within a month. The countermeasure is to bet a mix of markets that come out of the same model: spreads on the same game, alt totals, sometimes the over when the model genuinely likes it (rarely on MNF, but it happens — Bengals-Chiefs on a slow-week defensive resurgence might price the total too low).
The /build/new workflow makes this straightforward because the model outputs cover probability and total-under probability from the same feature pack. You wager off the highest-EV pick on each game, not off a fixed market bias. The bet mix will look natural to the book's risk team.
Bankroll sizing per pick
Use a fractional Kelly clipped at 2-3% of bankroll per pick. The MNF totals signal is too noisy to size aggressively. If the model's predicted under probability is 0.62 against a market-implied 0.50, your edge is 12 percentage points and Kelly says roughly 6% of bankroll — clip that to 2-3% so a bad variance week does not crater you.
Closing line value as the diagnostic
Track CLV game by game. A model that picks unders and watches the closing total tick down 0.5-1 point on the games it likes is showing real edge. A model that picks unders and watches the line stay flat or move up is probably losing on noise. After 30-40 MNF bets, the CLV distribution will tell you more than the ATS record. We cover the methodology in the CLV explainer.
Beyond MNF: where the same approach generalizes
The MNF distortion is the strongest instance of a broader pattern: any nationally televised, narrative-heavy game where casual money concentrates in one direction is a candidate for a similar pricing inefficiency. SNF totals have a weaker version of the same edge (the public still bets unders, but the inefficiency is smaller because Sunday already has the entire afternoon slate competing for attention). Thursday Night Football used to have a strong under edge driven by short rest, but the league has spread out the TNF rest schedule over the last three years and the edge has compressed.
International games (London, Munich, Frankfurt, São Paulo) are the other interesting bucket. The 2026 schedule has six international games, all in the early-season window. International games have historically gone under at a 58% clip — partly weather (London in October), partly travel fatigue (the team flying east faces a circadian penalty), partly broadcast-window oddities. The same MNF totals brick scores international games as a useful side application; add a "long-haul travel" binary input and the model picks up the signal.
Putting it on the /leaderboards
The community models on the leaderboards include several Monday-night specialists. The top public MNF totals model as of mid-May 2026 had a 2025 CLV of +0.34 points per bet across 18 wagers — small sample, but the right direction. Comparing your private MNF brick against the public leaderboard winner is the cheapest way to know whether your features matter or you are just rediscovering what the consensus already prices in. If your model agrees with the leaderboard 80% of the time, you have the same edge they have. If your model picks differently on the 4-5 highest-total games of the year, that disagreement is where the alpha lives.
Open a Monday-night totals brick in workshop next time you want to test a new feature — wind data, defensive funnel index, broadcaster lineup as a public-bias proxy. The infrastructure for the backtest is shared with every other model in the catalog, so the marginal cost of running another experiment is one click.
And if you publish your finished brick to the marketplace, other users can subscribe to the pick feed for the rest of the season. The MNF slot is small enough (17 regular-season games plus the Week 1 second slot) that even a casual subscriber count of 50-100 is reasonable for a tightly-scoped niche brick.
Price examples and pass rules
Use names as evidence, not decoration. The useful SEO win is that Josh Allen, Ja'Marr Chase, Bijan Robinson and Puka Nacua and Chiefs, Bills, Ravens, Eagles and Lions appear inside decisions, thresholds, and internal links instead of being dumped into a keyword list.
- Spread example: if Chiefs-Broncos opens Chiefs -3.5 and your fair number is -2.8, +3.5 is the bet, +3 is a pass, and the moneyline needs roughly +155 or better before it replaces the spread.
- Total example: if a Bills outdoor total opens 46.5 and wind moves from 8 mph to 21 mph, an under projection at 42.8 still needs a playable number; under 45 or better is different from chasing 43.5.
- Futures example: Bengals AFC North +280 is 26.3% before hold. If your fair number is 30%, stake modestly, track portfolio correlation, and avoid stacking every Burrow, Chase, and Higgins bet into the same thesis.
- CLV rule: a good write-up is not enough. Track whether the spread, total, prop, or futures price closed better than your entry before grading the process.
Use closing-line value guide, vig and hold guide, bet tracking workflow to keep the examples attached to measurable prices.
Research note board
Use this table to turn the guide into a decision note. The point is to know when the idea is actionable and when it is only context.
| Angle | Input to verify | Example application | Pass when |
|---|---|---|---|
| Market price | Spread, total, moneyline, prop price, or futures hold | Chiefs and Bills compared through PPR | The price has moved past the number that created the edge |
| Football or sport context | Role, pace, weather, injury status, opponent style | Josh Allen role news mapped to the relevant market | The original input changes or remains unconfirmed |
| Review loop | Entry, close, result, and reason code | closing line value logged with a clear thesis | You cannot explain whether the process beat the market |
Bet responsibly — set limits, never chase losses.
Average total points by weather bucket
Average combined points scored in NFL games by weather bucket over recent seasons. Wind above 20mph and snow each clip totals by 6-8 points vs domed games, which is why books move totals aggressively when forecasts shift.
NFL ATS cover-margin distribution
Distribution of (final margin − closing spread) across an NFL season. Roughly normal with mean ≈ 0 and standard deviation ≈ 13 points, which is why most ATS edges live in the ±1.5 point window.
