Published on Tue Oct 27 2026 20:00:00 GMT+0000 (Coordinated Universal Time) by cresencio
NFL Week 7 Recap: Eight Winners, a Difficult Week Against the Spread
Eight correct winners did not make this a strong week across the board. Our ensemble finished Week 7 8–6 on outright selections, 4–9–1 against the spread and 7–7 on totals. Chicago opened with a three-market sweep, Seattle handed us three Sunday-night misses, and Dallas closed Monday with a correct winner and cover—but a missed under.
All 14 games are final. This recap grades the forecasts issued October 20 at 3:07 p.m. CDT, using their original lines. The Week 7 preview preserves what we picked before kickoff.
The complete Week 7 record
| Prediction | Correct | Incorrect | Pushes | Correct rate |
|---|---|---|---|---|
| Outright winner | 8 | 6 | 0 | 57.1% |
| Against the saved spread | 4 | 9 | 1 | 30.8% |
| Saved game total | 7 | 7 | 0 | 50.0% |
Spread accuracy excludes the push: four correct selections from 13 decisions, not 14. Green Bay +3 pushed when Detroit won by exactly three. These are forecast-selection records, not returns from the filtered betting recommendations or a claim that every pick was wagered.
The ensemble’s season winner record is now 63 of 106 decisions, or 59.4%. There have been 107 completed games; the season’s one tie is excluded from that binary denominator. The season comparison retains the original published Week 4 forecast rather than substituting its later alternate version.
Every game against the original picks
Projected and final scores read away–home, matching the matchup. The projected scores come from the separate XGBoost margin and total models; the winner, spread and total selections come from their respective probability forecasts. W means correct, L incorrect, and Push neither. LAR means the Rams. Pittsburgh–New Orleans was played at a neutral venue with New Orleans designated home.
| Game | Projected score | Final score | Winner pick | Saved spread pick | Saved total pick |
|---|---|---|---|---|---|
| NE at CHI | 23.8–26.1 | 24–30 | CHI / W | CHI −3 / W | Over 50.5 / W |
| PIT vs. NO | 23.1–22.9 | 16–10 | PIT / W | NO +3.5 / L | Under 44.5 / W |
| SF at ATL | 26.7–24.3 | 20–29 | SF / L | SF −3.5 / L | Over 50.5 / L |
| CIN at BAL | 25.2–28.0 | 35–13 | BAL / L | BAL −3 / L | Over 51 / L |
| TB at CAR | 20.2–26.1 | 43–13 | CAR / L | CAR −3 / L | Over 44.5 / W |
| NYG at HOU | 17.9–22.7 | 27–21 | HOU / L | HOU −9.5 / L | Under 43 / L |
| IND at MIN | 20.8–22.9 | 10–16 | MIN / W | MIN −6.5 / L | Over 41.5 / L |
| MIA at NYJ | 18.3–23.1 | 28–25 | NYJ / L | NYJ −5 / L | Over 39.5 / W |
| CLE at TEN | 20.4–21.0 | 27–24 | CLE / W | TEN +2 / L | Over 38.5 / W |
| DEN at ARI | 20.7–22.1 | 30–16 | DEN / W | DEN −4 / W | Over 43.5 / W |
| GB at DET | 26.0–30.7 | 17–20 | DET / W | GB +3 / Push | Over 53.5 / L |
| LAR at LV | 23.9–24.8 | 35–21 | LAR / W | LAR −4.5 / W | Over 47.5 / W |
| KC at SEA | 24.2–21.7 | 3–20 | KC / L | KC +2.5 / L | Over 42.5 / L |
| DAL at PHI | 22.7–22.1 | 38–21 | DAL / W | DAL −4.5 / W | Under 44.5 / L |
Finals were reconciled in Tuesday’s closeout against the official NFL Week 7 results. Score projections are rounded independently from the saved full-precision values.
Houston was the biggest confidence miss. Its 73.7% winner forecast led the board, but the Giants won 27–21. Houston −9.5 and Under 43 also missed. Carolina (65.6%) and San Francisco (64.3%) lost as well, leaving three of our four strongest winner selections incorrect. An upset is possible at any probability below 100%; this cluster deserves attention without treating one week’s results as a complete calibration test.
Minnesota shows why winner and spread accuracy need separate accounting. The Vikings won 16–10, delivering the 67.9% outright pick. Their six-point margin fell half a point short of covering −6.5. Detroit’s 20–17 win delivered its winner pick but only a push for the separate Green Bay +3 selection.
Denver and the Rams joined Chicago as the week’s three complete sweeps. Denver’s 30–16 win covered −4 and cleared 43.5; the Rams’ 35–21 win covered −4.5 and cleared 47.5. Yet the separate point model narrowly favored Arizona and Las Vegas. Getting all three probability-based selections right did not make those score projections right.
Dallas delivers the side, overshoots the score
Philadelphia briefly led 21–20 after opening the second half with an 84-yard touchdown drive. Dallas answered with Dak Prescott’s touchdown pass to CeeDee Lamb and a two-point conversion, beginning an 18–0 finish. Prescott threw three touchdown passes without an interception, while Javonte Williams rushed for 155 yards and a touchdown. Philadelphia recorded no sacks or quarterback hits. Official Eagles recap.
The 38–21 final made Dallas to win at 55.5% and Dallas −4.5 correct. The Cowboys covered the original handicap by 12.5 points. Their 59 combined points defeated Under 44.5, finishing 14.5 above that threshold. Monday’s BetMGM total had moved to 44, but the published selection and its 52.5% probability remain attached to 44.5.
The Monday preview also published a near-tie point projection. Here is how far the final moved from it:
| Point-model measure | Frozen projection | Actual |
|---|---|---|
| Dallas points | 22.65 | 38 |
| Philadelphia points | 22.09 | 21 |
| Dallas winning margin | 0.56 | 17 |
| Combined points | 44.74 | 59 |
Dallas scored 15.35 more points than projected. The total projection was 14.26 points too low, and the margin missed by 16.44. Those errors use the full-precision saved values before rounding. Philadelphia finished near its projected score; the larger miss was Dallas’s offense.
The preview’s unresolved availability also has an answer: Saquon Barkley, Jalyx Hunt and Jonathan Greenard were active. Barkley gained 62 yards and scored on seven carries, with limited snaps. Their availability did not prevent Dallas from controlling the closing stretch; it also does not establish how much any injury changed the score or betting line. Eagles final inactives, game recap.
Across the week’s three primetime games, we went 2–1 on winners, 2–1 on spreads and 1–2 on totals. Chicago’s 30–24 win delivered Chicago −3 and Over 50.5. Seattle’s 20–3 win defeated Kansas City to win, Kansas City +2.5 and Over 42.5; it also ended the earlier winner and spread streaks. The point model had projected 45.98 total points in Seattle, 22.98 too many. Monday’s correct side does not erase Sunday’s misses.
Which models handled the week best?
Bayesian led winner accuracy in this 14-game sample. The weighted ensemble matched XGBoost and Random Forest, finishing behind Bayesian and Elo.
| Winner model | Correct–incorrect |
|---|---|
| Bayesian | 10–4 |
| Elo | 9–5 |
| XGBoost | 8–6 |
| Random Forest | 8–6 |
| Weighted ensemble | 8–6 |
| Logistic regression | 7–7 |
The ensemble’s log loss was 0.72576 and Brier score 0.26410. Both measure probability error, with lower scores better. For context, assigning every game a 50% win probability produces 0.69315 log loss and 0.25000 Brier. The ensemble beat a 50% winner hit rate this week but performed worse than that neutral probability benchmark on both measures. Winner accuracy alone misses that distinction.
The trained spread and total models also differed:
| Market model | Saved spread record | Saved total record |
|---|---|---|
| Logistic / Linear | 7–6–1 | 8–6 |
| XGBoost | 5–8–1 | 5–9 |
| Random Forest | 5–8–1 | 5–9 |
| Ensemble aggregate | 4–9–1 | 7–7 |
The first row uses the Logistic spread model and Linear total model. Elo and Bayesian market proxies are diagnostic, not independently trained spread or total models, and are excluded from this comparison. A one-week lead does not establish lasting superiority or betting profitability.
The probabilities remain raw and unvalidated. Dallas’s saved 57.6% chance to cover −4.5 exceeded its 55.5% chance to win outright—an inconsistency between separately fitted models that Monday’s correct result does not repair. The refreshed monitoring assessment was still pending when these recap notes were prepared; an older GREEN status is not a new endorsement of Week 7.
Week 8 gets its own preview when the new forecasts are ready. This week leaves a clear set of results to carry forward: eight correct winners, nine incorrect spread selections, and score projections that struggled with both Seattle’s low-scoring win and Dallas’s 38-point finish.
Source note: This review uses the October 27 Week 7 closeout and verified original-line grades, the unchanged October 20 forecasts, and the season summary through Week 7. Football details added to the results notes are linked to official team reporting. Later market quotes do not rewrite the original selections, and none of these records represents portfolio returns.
Written by cresencio
← Back to blog
