NFL Week 10 AI Predictions, Graded: Road Teams Went 10-4 and Home Field Disappeared
Every week this NFL season, we're pitting 8 AI agents (4 with Daggy's statistical engine and 4 without) against Kalshi's prediction market. Here's the Week 10 recap, with every AI prediction graded against the final score. Live forecasts for every game are on our NFL page.
NFL Week 10 AI Predictions vs Kalshi: Brier Scores
Road teams won 10 of 14 games in Week 10. Daggy and Kalshi both picked 9 of 14 winners, and Daggy edged Kalshi on Brier score, 0.194 to 0.196 (lower is better). The closing betting line scored 0.196 too.
The margin is small, and so was the gap in what the two expected. Before kickoff, our average forecast gave the home team a 43.7% chance, and Kalshi gave it 44.2%. Both expected a road-leaning week. The surprise was which road teams won.
For the season, Kalshi leads 0.216 to 0.219 in Brier and 100-50 to 98-52 picking winners.
NFL Week 10 AI picks vs. results: every game
Probabilities are each forecaster's chance for the team it picked, made before kickoff. "Extremized" pushes Daggy's forecast away from 50-50 (explained in Week 9).
| Game | Daggy pick | Kalshi pick | Final | Daggy Brier | Kalshi Brier | Extremized Brier |
|---|---|---|---|---|---|---|
| Commanders at Giants | Giants 64% | Giants 56% | Commanders 24-21 | 0.406 | 0.314 | 0.528 |
| Patriots vs Lions (Munich) | Lions 61% | Lions 59% | Lions 35-10 | 0.153 | 0.168 | 0.100 |
| Chiefs at Falcons | Chiefs 71% | Chiefs 63% | Chiefs 35-13 | 0.084 | 0.137 | 0.031 |
| Texans at Browns | Texans 63% | Texans 66% | Browns 27-24 | 0.397 | 0.436 | 0.512 |
| Vikings at Packers | Packers 60% | Packers 60% | Vikings 30-10 | 0.366 | 0.360 | 0.458 |
| Dolphins at Colts | Colts 65% | Colts 66% | Colts 24-22 | 0.123 | 0.116 | 0.066 |
| Panthers at Saints | Saints 50% | Saints 56% | Panthers 20-17 | 0.251 | 0.314 | 0.252 |
| Bills at Jets | Bills 79% | Bills 79% | Bills 31-12 | 0.045 | 0.044 | 0.009 |
| Jaguars at Titans | Jaguars 80% | Jaguars 83% | Jaguars 40-23 | 0.039 | 0.029 | 0.007 |
| Rams at Cardinals | Rams 75% | Rams 80% | Rams 34-6 | 0.061 | 0.040 | 0.016 |
| Seahawks at Raiders | Seahawks 74% | Seahawks 72% | Seahawks 21-13 | 0.070 | 0.078 | 0.021 |
| 49ers at Cowboys | 49ers 70% | 49ers 65% | 49ers 46-40 | 0.091 | 0.122 | 0.036 |
| Steelers at Bengals | Bengals 57% | Bengals 63% | Bengals 38-18 | 0.186 | 0.137 | 0.146 |
| Chargers at Ravens | Ravens 67% | Ravens 67% | Chargers 23-20 | 0.447 | 0.449 | 0.595 |
| Week 10 | 9-5 | 9-5 | 0.194 | 0.196 | 0.198 |
Where did home field advantage go?
Home teams have won 71 of 150 games this season (47.3%). From 2000 to 2025 they won 56%. Week 10 was the extreme case: road favorites went 6-1, and home favorites went 3-4.
Kalshi made seven home teams favorites. The Lions, Colts and Bengals won. The Giants, Packers, Saints and Ravens lost, the Ravens and Giants by three points each. Of the seven road favorites, only Houston lost, 27-24 in Cleveland.
The AI agents had already stopped trusting home field
Our agents keep notes between weeks, including what they have learned about the league. Going into Week 10, several had written down that home field was worth less this season, and their forecasts said so:
- Opus without the engine started most games at 50%, not the usual home-team base rate: "Home teams went 50-53-1 in Weeks 3-9, so home field has been worth little this season."
- Kimi with the engine, on Commanders at Giants: "What held me down: 2026's home-field advantage is essentially dead." It had the Giants at 60%. Washington won.
- Kimi with the engine again, on 49ers at Cowboys, priced it as "roughly SF as a ~4-point road favorite … in a season where home-field edge is running near zero." It gave Dallas 22%. San Francisco won.
The agents also play neutral-site games differently. The Lions were the designated home team in Munich, and seven of the eight agents treated it as a neutral site. Kimi with the engine: "My league card records 2026 international designated-home teams at ~1-6 with maybe no crowd, so I zeroed out home-field advantage." Detroit won 35-10.
Where we were more confident than Kalshi on a road favorite, it paid off: the Chiefs (71% vs 63%) and 49ers (70% vs 65%) both won. Leaning away from the home team didn't work everywhere, though. We had the Bengals at 57% to Kalshi's 63%, and they won by 20.
Is the home field decline real, or noise?
After Week 10, two agents with the engine tested that question against 26 years of games, and they don't agree.
Kimi with the engine found a long decline. In its notes, home teams won 57.6% in the earliest five-year stretch since 2000, 53.9% in the most recent, and 47.3% this season.
Opus with the engine tested whether a home team's shortfall in the first nine weeks predicts the rest of the season. It found no relationship (a correlation of −0.05 across 26 seasons), and wrote that trimming home teams for it had cost it in Cincinnati: "Weeks 1-9 home residual does not predict weeks 10+ … yet I trimmed home teams for it and lost."
Both can be true: home field advantage has been shrinking for years, and this season's 47% is probably partly bad luck. The agents will carry both notes into Week 11.
Chargers vs Ravens AI prediction: the costliest miss
All eight agents picked Baltimore, and so did Kalshi (67%). The winless Chargers (0-8) won 23-20.
Baltimore's rating was built on Lamar Jackson, who is on IR. Kimi with the engine spotted it: "BAL's Elo is Lamar-stale — Jackson is on IR after thumb surgery, and Elo can't see that." It still had Baltimore at 62%, because Jacoby Brissett was a clear upgrade over the third-stringer who had started. Brissett had about a week in the system. In the post-mortem, the blind supervisor wrote a new rule: a rating is stale "if the rating came from an injured starter and the team has lost 4+ straight since."
Did extremizing help this time?
No. In Week 9, extremizing would have won the week, 0.081 to Kalshi's 0.117. In Week 10 it would have scored 0.198, worse than both our forecast and Kalshi. It was better on the 9 games we got right and worse on all 5 we got wrong.
Over Weeks 4-10 it now trails the plain average, 0.212 to 0.211. Extremizing is a bet that the favorites keep winning, and in Week 10 five of them didn't.
Did the agents with the engine win again?
No. For the first time this season, the four agents without Daggy's engine did better, by 0.004 on Brier. Of 56 head-to-head pairs, the engine won 24, lost 28 and tied 4. The gap is well inside the noise (95% interval −0.010 to +0.016).
| Week | Pairs | Engine minus no-engine (Brier, lower = engine better) | Engine won the pair |
|---|---|---|---|
| 3 | 64 | −0.010 | 53% |
| 4 | 64 | −0.011 | 53% |
| 5 | 60 | −0.012 | 55% |
| 6 | 56 | −0.000 | 50% |
| 7 | 56 | −0.013 | 59% |
| 8 | 56 | −0.018 | 59% |
| 9 | 60 | −0.003 | 52% |
| 10 | 56 | +0.004 | 43% |
One of the engine's best games was Jaguars at Titans. Opus with the engine had Tennessee at 13%, Opus without it at 26%. Jacksonville won by 17. One of its worst was Chiefs at Falcons, where Gemini with the engine gave Atlanta 34% against 21% without it.
How our AI NFL predictions work
- Eight AI agents per game. Four models (Kimi K3, Claude Opus 5.5, GPT-6 Sol and Gemini 3.8 Flash), each run twice: once with Daggy's statistical engine and once without.
- Blind to the betting market. The agents search the web for news and injuries, but betting lines and odds are withheld so they can't copy the market (why).
- They learn between weeks. Every agent writes a post-mortem on each game and keeps notes. The agents with the engine test their notes against 26 years of NFL data.
- One data-driven forecast. The eight probabilities are averaged into Daggy's prediction, made hours before kickoff and never changed.
- Graded every week against Kalshi's prediction market with the Brier score.
Top questions, answered
How accurate were the AI NFL predictions in Week 10? Daggy's agents picked 9 of 14 winners with a Brier score of 0.194. Kalshi's market also picked 9 of 14, with 0.196.
Is NFL home field advantage gone in 2026? Home teams have won 47.3% of games this season, against 56% from 2000 to 2025, and went 4-10 in Week 10. Home field advantage has been shrinking for two decades, but a season this low is likely partly noise.
Which AI model was most accurate in Week 10? Kimi K3 (0.192 Brier), then Gemini 3.8 Flash (0.193), Claude Opus 5.5 (0.197) and GPT-6 Sol (0.202).
Previous: Week 9: 14-1, and the cost of hedging · All roundups