Week 1 underdogs went 218-196-10 ATS from 1999 through 2025. That is a 52.7% cover rate across 414 non-push decisions. It is a real edge-shaped number. It is also 27 seasons compressed into one row, which is how a useful note turns into a blind card before lunch.
The better read keeps the splits attached. Road dogs went 143-124-8. Home dogs went 75-72-2. Dogs getting seven or more points went 47-29. The smallest dogs were close to even. The record is worth inspecting because Week 1 is noisy. It is not a permission slip to stop pricing the game.
The complete Week 1 dog record
The sample contains 424 non-pick’em Week 1 games. Underdogs covered 218, failed to cover 196, and pushed 10. They also won 144 games outright. The outright wins are fun. The ATS ledger is the grade.
For comparison, underdogs in every other regular-season week went 3,232-3,102-179 ATS, a 51.0% cover rate. Week 1’s 52.7% rate was about 1.6 percentage points higher. The difference exists in this snapshot, but it is not a canyon. One should not market a modest gap as a separate law of football.
The recent five-season slice was stronger. Week 1 dogs went 44-36 with no pushes from 2021 through 2025, a 55.0% cover rate over 80 decisions. That window is useful as a persistence check. It is also much smaller than the full sample and was selected after the broad angle was already visible.
Road dogs led home dogs
| Week 1 dog location | ATS record | Cover rate | Outright wins |
|---|---|---|---|
| Road underdog | 143-124-8 | 53.6% | 94 |
| Home underdog | 75-72-2 | 51.0% | 50 |
| All underdogs | 218-196-10 | 52.7% | 144 |
Road underdogs carried most of the sample and most of the covers. They had 275 games versus 149 for home dogs. The 2.6-point rate gap is descriptive. It does not isolate travel, crowd, or scheduling. It simply says which side beat the closing number more often in the selected Week 1 rows.
The home-dog record is another warning against a clean slogan. “Week 1 dogs cover” sounds broad. Home dogs were only 75-72-2. Remove the road side and the dramatic headline becomes a near-even record.
The spread buckets
| Points received | ATS record | Cover rate | Outright wins |
|---|---|---|---|
| +1 to +3 | 86-82-8 | 51.2% | 77 |
| +3.5 to +6.5 | 85-85-2 | 50.0% | 49 |
| +7 or more | 47-29-0 | 61.8% | 18 |
The big-dog bucket did the work. Teams getting seven or more points covered 61.8% of 76 games. The middle bucket was exactly 85-85 after excluding two pushes. Small dogs were 86-82-8. If the +7 group is removed, the broad Week 1 angle becomes much less exciting.
That does not make +7 an automatic trigger. Seventy-six games spread across 27 seasons is fewer than three per season on average. The article did not search every possible cut at quarter-point resolution. It used three declared, readable buckets. A live strategy would still need to lock the rule, include the actual price, and test new seasons without moving the goalposts.
The annual path is messy
The full record was not a steady drip of covers. Week 1 dogs went 11-3 in 1999, 6-10 in 2008, 11-5 in 2014, 12-4 in 2021, and 7-9 in each of 2024 and 2025. Those examples are not a curated “system.” They show how a 16-game annual sample can swing.
Across the 27 seasons, a single season contributes only 14 to 16 non-pick’em decisions in most years. Reversing only a small number of decisions can move an annual percentage sharply. The long sample is the evidence. The annual rows are a reminder that the path to the aggregate was uneven.
Why Week 1 invites bad inference
There is no completed current-season record before Week 1. That makes historical angles feel heavier than they are. Every bettor has the same blank standings page, so a 27-season split can sound like privileged information. It is public history graded at public closes.
The source also contains no preseason games. It cannot tell us how a team’s preseason performance translated into Week 1. It contains no opening spread, so it cannot tell us whether an underdog carried a different number earlier than it did at close. It contains the closing number and final score. That is enough for a clean ATS record and not enough for the story around it.
My old move was to fill those gaps with confidence. The disciplined move is to label them. Missing opening data is not “no movement.” Missing preseason rows are not “preseason did not matter.” They are unavailable questions in this snapshot.
The Donk Check
The record earns a lean for further research, not a blind card. Week 1 dogs covered 52.7%. Road dogs were better than home dogs. The +7-or-more bucket was much better than the two smaller buckets. Those are the receipts.
The next current decision still starts with the posted spread. Check the NFL picks page for a graded pick tied to that number. Use Analytics to test the Week 1 filter without changing the date window after seeing the answer. Read what ATS means if the difference between 218-196-10 and 52.7% is not automatic yet.
How I would test the angle next
I would register one rule before the next completed season: regular-season Week 1, non-pick’em underdogs, graded at a named closing source. I would keep road versus home and the three spread buckets as diagnostics, not as a pile of new strategies. Then I would append the season without editing the old record.
I would also retain the actual side price. The W-L-P line is comparable across the archive, but a return calculation needs the price paid. I would not assume every spread was offered at the same price. And I would report any no-data state directly rather than treating an unplayed 2026 schedule row as a loss, win, or empty record.
Why the +7 bucket needs a holdout
The large-dog row is the strongest number in the article and therefore the easiest one to overfit. It has 47 covers and 29 losses. A few reversed decisions would pull the rate toward the other buckets. More important, the bucket was observed in the same archive used to describe it.
The next test should not split +7 into new subgroups until one looks best. It should carry the declared +7-or-more rule into new completed Week 1 games. If the result weakens, that is evidence. If it persists, the record gains decisions without changing its definition.
A holdout also needs the exact closing source and side price. Otherwise a future analyst can choose the friendliest close after the game. The historical CSV gives one stored close; a live system should timestamp the source it grades.
The same holdout rule applies to the location split. Road dogs led the full sample, but road should remain a diagnostic unless it was declared before the next test. Promoting every favorable subgroup to a separate pick rule multiplies chances to find noise and then call it signal.
Where these numbers come from
How we counted: We concatenated the four NFL CSV shards, kept played rows with game_type = REG, seasons 1999–2025, week = 1, and a nonzero closing spread_line. The underdog is the side opposite the favorite implied by the dictionary’s spread sign. Underdog ATS margin is the negative of favorite ATS margin; positive is a cover, zero a push, negative a loss. Points received are abs(spread_line). The comparison with other weeks uses the same rules with week != 1. Cover rate excludes pushes. Source files live under data/training-snapshots/nfl/; definitions and shard hashes are in DICTIONARY.json and manifest.json.
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.
Model calibration: predicted vs observed
Predicted win probability bucket vs the empirical win rate inside that bucket on the test set. Points on the y=x reference line are perfectly calibrated; points below mean the model is overconfident in that bucket.
