Fourthwise’s Early model has made its Week 4 calls. Now we’re putting those predictions on trial.

NFL predictions are easiest when everything agrees. The model likes one team. Recent performance agrees. The injury report cooperates. The market points in the same direction.

Week 4 is not that simple.

Across the remaining slate, there are games where the Fourthwise model and the surrounding evidence line up almost perfectly. There are others where the model is going directly against the market. And there are a couple where developments since the prediction cutoff give us enough reason to challenge the machine.

That is exactly what we want to investigate.

The predictions discussed below are Fourthwise’s Early predictions, published before the final pregame window. The current Early model’s historical evaluation covers 271 games, with 64.2% winner accuracy, 0.643 log loss and a 0.226 Brier score. fourthwise-week4-article-data

Roughly four hours before kickoff, Fourthwise transitions to its Final prediction, where the stronger CatBoost model gets the latest pregame information. So this article is deliberately a snapshot of the argument before that final update.

We’ll start with the model, investigate the matchup, injuries, recent performance and market context, and then make the call:

Trust the model — or override it?

See the live Fourthwise Week 4 predictions

Indianapolis Colts at Washington Commanders

Fourthwise Early prediction: Washington 53.8% — Indianapolis 46.2%

This might be the most interesting disagreement on the entire board.

Fourthwise narrowly favors Washington, but underneath that headline prediction is a warning sign: the model explanation’s pregame team-strength factor actually leans toward Indianapolis. fourthwise-week4-article-data

The market disagrees with the published prediction even more strongly. Fourthwise’s nine-book snapshot has Indianapolis around -4 and -210, with Washington +172. fourthwise-week4-article-data

And then comes the biggest development of all.

Washington will not have Jayden Daniels. The Commanders ruled him out with his elbow injury, leaving Marcus Mariota to start in London. Indianapolis has its own problems — Keenan Allen was downgraded to out Saturday — but the quarterback change carries substantially more weight.

The teams are both 1-2, and Fourthwise’s recent-form comparison is almost neutral: Washington has averaged a -5.7 scoring margin compared with Indianapolis at -6.3. fourthwise-week4-article-data

That isn’t enough to overcome the quarterback situation and the broader disagreement between the Early prediction and the market.

Decision: OVERRIDE

Our pick: Indianapolis Colts

This is exactly the kind of matchup where the Final model will be fascinating. The Early prediction says Washington. The evidence available closer to kickoff increasingly points the other way.

Tennessee Titans at Baltimore Ravens

Fourthwise Early prediction: Baltimore 81.2% — Tennessee 18.8%

Not every prediction requires an investigation worthy of a courtroom drama.

Baltimore owns one of the strongest probabilities on the entire Fourthwise Week 4 board. Even with injuries affecting the Ravens, the gap here is enormous.

The more important question isn't whether Baltimore has vulnerabilities. Every team does. It is whether Tennessee presents enough evidence to justify ignoring an 81% model probability.

We don't see it.

Baltimore has enough offensive firepower to force Tennessee into a game script the Titans may struggle to match. Lamar Jackson, Derrick Henry and Zay Flowers give the Ravens multiple ways to create explosive plays, and nothing in the broader Week 4 picture provides enough reason to overturn the model.

Decision: TRUST

Our pick: Baltimore Ravens

New York Jets at Chicago Bears

Fourthwise Early prediction: Chicago 73.7% — New York 26.3%

The probability looks decisive. The personnel situation makes it less comfortable.

Chicago enters without Caleb Williams, which immediately increases uncertainty at the most important position on the field. The Jets also arrive with meaningful availability problems of their own.

So the question becomes whether Chicago's underlying advantage survives the quarterback downgrade.

For now, yes.

The Bears are coming off a convincing 27-7 win over Philadelphia, while the Jets haven't supplied enough evidence to justify flipping a model that gives Chicago nearly three chances in four.

But this belongs in a different category from Baltimore-Tennessee.

Decision: TRUST — with caution

Our pick: Chicago Bears

This is one to revisit when the Final prediction arrives.

Arizona Cardinals at New York Giants

Fourthwise Early prediction: New York Giants 57.1% — Arizona 42.9%

The Giants don't have a huge cushion here, but the model gives them a meaningful one.

This matchup looks more like a modest home lean than a game where one statistical edge dominates the analysis. Arizona has defensive availability concerns, while New York enters without the kind of official game-status injury cluster that would force us away from the Early prediction.

At 57.1%, Fourthwise isn't claiming certainty.

Neither are we.

Decision: TRUST

Our pick: New York Giants

Jacksonville Jaguars at Cincinnati Bengals

Fourthwise Early prediction: Jacksonville 53.6% — Cincinnati 46.4%

Now we get another genuine disagreement.

Both teams enter 2-1, and Cincinnati is the home team. The market also favors the Bengals.

But Fourthwise says Jacksonville.

The strongest argument for staying with the model is recent performance. In the production evidence behind the prediction, Jacksonville's average scoring margin is +15.3, compared with Cincinnati at +5.7 — a strong 9.7-point recent-form advantage for the Jaguars.

Cincinnati still has the quarterback capable of blowing up any prediction. Joe Burrow entered the week playing at a high level, and that alone prevents this from becoming a confident Jacksonville call.

But Jacksonville's start has not been accidental. The Jaguars have combined strong recent results with impressive red-zone performance on both sides of the ball.

The model is taking the less obvious side.

We're staying with it.

Decision: TRUST

Our pick: Jacksonville Jaguars

This is one of our favorite model-vs-market tests of Week 4.

Green Bay Packers at Tampa Bay Buccaneers

Fourthwise Early prediction: Green Bay 56.5% — Tampa Bay 43.5%

The Early model already preferred Green Bay.

Then Tampa Bay lost Baker Mayfield.

Mayfield is out with a dislocated right thumb, and undrafted rookie Jalon Daniels will make his first NFL start. NFL

That doesn't make Green Bay bulletproof.

The Packers have serious problems of their own, particularly along the offensive line and in a rushing attack that entered Week 4 at the bottom of the league. NFL.com's Week 4 analysis noted Green Bay had produced only 146 rushing yards through three games. NFL

Tampa's defense therefore has a plausible path to keeping this ugly.

But asking a rookie quarterback to make his first start against a Green Bay team the model already preferred is enough for us to stay put.

Decision: TRUST

Our pick: Green Bay Packers

Los Angeles Rams at Philadelphia Eagles

Fourthwise Early prediction: Philadelphia 60.8% — Los Angeles 39.2%

Here is the biggest challenge to the Fourthwise model this week.

And unlike Indianapolis-Washington, this isn't a near coin flip.

Fourthwise gives Philadelphia a legitimate 60.8% win probability.

Yet almost everything that has happened around the prediction deserves attention.

The market has Los Angeles favored on the road. Recent scoring margin also favors the Rams: Philadelphia sits at -4.7 compared with Los Angeles at -0.7.

Then look at Philadelphia's injury report.

DeVonta Smith, Dallas Goedert, Marquise Brown, Zack Baun, Marcus Epps and Fred Johnson are all out. The Rams have important losses too — most notably Aaron Donald — but Puka Nacua is returning after missing two games. NFL

Philadelphia is also coming off a 27-7 loss to Chicago.

There is an argument for trusting the model anyway. The Eagles are at home, and the Fourthwise prediction is not a marginal 51-49 call. Its pregame team-strength factor strongly favors Philadelphia.

But this exercise isn't supposed to rubber-stamp the algorithm.

When enough important information accumulates on the other side, we should be willing to disagree.

This is that game.

Decision: OVERRIDE

Our pick: Los Angeles Rams

If Fourthwise's Final CatBoost prediction remains on Philadelphia after processing the latest pregame environment, this becomes an even more interesting disagreement.

New England Patriots at Buffalo Bills

Fourthwise Early prediction: Buffalo 68.2% — New England 31.8%

Buffalo doesn't need much help from us here.

The Bills have the stronger quarterback, the stronger overall offensive ceiling and home field. New England also enters the matchup with meaningful defensive absences.

A 68.2% probability still leaves plenty of room for an upset — roughly one outcome in three according to the model — but we're looking for evidence strong enough to overturn the favorite.

There isn't enough.

Decision: TRUST

Our pick: Buffalo Bills

Dallas Cowboys at Houston Texans

Fourthwise Early prediction: Houston 61.2% — Dallas 38.8%

This is where process matters.

Houston enters 0-3. Dallas is 1-2. If you reduced the matchup to record alone, Houston at 61.2% would look strange.

But predictions shouldn't be standings tables.

Houston's defense has shown the ability to generate pressure, while Dallas presents a much tougher offensive test than Houston's record might suggest. The concern is whether the Texans can turn enough possessions into points; their offensive efficiency in scoring territory has been a major early-season problem.

So we're not treating 61.2% like a lock.

But we also don't have enough evidence to discard it.

Decision: TRUST — with caution

Our pick: Houston Texans

Miami Dolphins at Minnesota Vikings

Fourthwise Early prediction: Minnesota 75.1% — Miami 24.9%

Minnesota is 3-0. Miami is 0-3.

The model sees the gap clearly.

The major counterargument is personnel: Minnesota will be without Justin Jefferson, removing one of football's most important offensive weapons.

That's significant enough to reduce our confidence.

It isn't significant enough to flip the game.

Minnesota has built enough of a performance advantage through three weeks that asking Miami to erase it solely because Jefferson is unavailable would require evidence we don't have.

Decision: TRUST

Our pick: Minnesota Vikings

Kansas City Chiefs at Las Vegas Raiders

Fourthwise Early prediction: Kansas City 63.2% — Las Vegas 36.8%

This might be the best game on the slate.

Both teams enter 3-0. Both have scored at a high level. And unlike several other matchups, recent form doesn't create a major separation: Fourthwise has Kansas City at an average +12.7 scoring margin and Las Vegas at +11.3.

So why stay with Kansas City?

Because when the margins narrow, quarterback play and overall offensive sustainability matter even more.

Patrick Mahomes and the Chiefs have produced the more convincing all-around offensive profile, while Kansas City's defense gives it another route to winning if the game doesn't turn into a shootout.

Las Vegas is good enough to make 63.2% uncomfortable.

That's not the same thing as having enough evidence to overturn it.

Decision: TRUST

Our pick: Kansas City Chiefs

Denver Broncos at San Francisco 49ers

Fourthwise Early prediction: San Francisco 51.6% — Denver 48.4%

This is effectively a coin flip.

And yet the supporting Fourthwise data are nowhere near as close as the headline probability.

San Francisco's recent average scoring margin is +16.0.

Denver's is -3.3.

That's a 19.3-point gap, categorized as a strong recent-form advantage in Fourthwise's contextual evidence.

The matchup analytics tell a similar story. San Francisco's offense has been substantially more efficient, while Brock Purdy enters the matchup with nine touchdown passes against one interception. NFL.com's game center lists Purdy at 789 yards, a 72.3% completion rate and that 9-to-1 TD/INT ratio through three games. NFL

There are reasons the model remains cautious. Denver is 2-1 and has shown it can survive difficult game scripts. San Francisco also has significant defensive injuries.

But this is a fascinating example of probability versus supporting evidence.

The model says the game is close.

The matchup evidence says San Francisco has played substantially better football.

We're not overriding it.

Decision: TRUST

Our pick: San Francisco 49ers

Los Angeles Chargers at Seattle Seahawks

Fourthwise Early prediction: Seattle 69.8% — Los Angeles 30.2%

This is one of the cleaner model calls.

Fourthwise's recent-form evidence gives Seattle an average scoring margin of +8.3 and the Chargers -10.7 — a massive 19-point separation.

The quarterback numbers reinforce the concern. Justin Herbert entered Week 4 with three touchdown passes, four interceptions, a 73.4 passer rating and negative EPA per dropback. NFL Fantasy

And the Chargers now have to take that offense into Seattle.

There are injury concerns on both sides. Los Angeles has several questionable contributors, while Seattle will be without Zach Charbonnet. NFL

But those aren't enough to erase what we've seen through three games.

Decision: TRUST

Our pick: Seattle Seahawks

Detroit Lions at Carolina Panthers

Fourthwise Early prediction: Detroit 65.6% — Carolina 34.4%

This game has one of the more interesting strength-on-weakness combinations of Week 4.

Carolina has demonstrated that it can move the ball through the air. Detroit's pass defense has been vulnerable.

That creates a legitimate path for the Panthers.

The problem is what happens on the other side.

Detroit can attack Carolina on the ground, control the structure of the game and generate pressure without needing to sell out defensively. Carolina also enters with important personnel losses.

The matchup therefore has upset ingredients without providing enough evidence to make the upset our primary expectation.

Decision: TRUST

Our pick: Detroit Lions

Atlanta Falcons at New Orleans Saints

Fourthwise Early prediction: Atlanta 52.1% — New Orleans 47.9%

We finish with another near coin flip.

And another model-market disagreement.

Fourthwise narrowly selects Atlanta. The market favors New Orleans.

The Saints have produced the better scoring offense through three games, while Atlanta's passing game has been inconsistent. But this matchup may be decided somewhere else entirely: on the ground.

Atlanta has been one of the league's most productive rushing teams, and New Orleans has struggled defending the run. On the other side, Atlanta's run defense has been excellent.

That gives the Falcons a plausible formula that doesn't require their passing game to dominate.

Fourthwise's recent scoring-margin evidence actually favors New Orleans, -0.7 to Atlanta's -5.7, which is another reason not to treat the 52.1% prediction as anything more than a lean.

Still, the matchup mechanics give us enough reason to stay on the model's side.

Barely.

Decision: TRUST — with caution

Our pick: Atlanta Falcons

So, How Often Did We Trust the Model?

Across the 15 remaining Week 4 games, we're staying with Fourthwise's Early prediction in 13 and overriding it in two.

The two disagreements are meaningful:

Indianapolis over Washington — Fourthwise has Washington at only 53.8%, the market strongly favors Indianapolis, and Jayden Daniels is out.

Los Angeles Rams over Philadelphia — Fourthwise gives Philadelphia a much stronger 60.8% probability, but the injury situation, market, recent performance and Puka Nacua's return collectively push us toward Los Angeles.

Those aren't declarations that the model is “wrong.”

They're the entire point of the exercise.

A probabilistic model answers one question:

Given the information and relationships captured by the model at its cutoff, which outcome is more probable?

Football analysis can then ask another:

What has changed, what might the model be missing, and is the new evidence strong enough to change our decision?

Sometimes the answer is yes.

Usually, this week, it isn't.

The Most Interesting Part Comes Next

These are the Early predictions.

Before kickoff, Fourthwise will replace them with the Final predictions, powered by the CatBoost model and the latest available pregame information.

That gives us an experiment we can actually follow.

We now have three layers:

Early Model → Evidence Review → Final Model

For Indianapolis-Washington and Rams-Philadelphia, we'll find out whether the Final model moves in the same direction as our analysis.

For games like Denver-San Francisco and Atlanta-New Orleans, we'll see whether close Early probabilities become more decisive — or flip entirely.

And for Baltimore-Tennessee, Minnesota-Miami and Seattle-LA, we'll see whether the strongest Early calls survive the final information window.

That's more useful than pretending any prediction is certain.

Check the live Fourthwise Predictions Hub before kickoff

Explore Fourthwise Matchup Analytics

Check the latest Fourthwise Injury Wire

Predictions are probabilistic estimates, not guarantees. Fourthwise content is for informational and entertainment purposes.