Published on Tue Oct 13 2026 17:25:20 GMT+0000 (Coordinated Universal Time) by cresencio
NFL Week 5 Results: Strong Winners, Weaker Spreads and Totals
Week 5 ended with Buffalo’s 37–34 overtime win at Los Angeles and a strong record for our outright selections: 11 correct, three incorrect and one tie. The spread and total predictions were less successful. Those results deserve their own accounting, rather than borrowing confidence from the winner picks.
All 15 games are complete. This review grades the predictions saved before the week began, keeping their probabilities and original spread and total lines intact. The Week 5 preview preserves the full pregame board.
The complete Week 5 record
| Prediction | Correct | Incorrect | Other result | Accuracy excluding ties or pushes |
|---|---|---|---|---|
| Outright winner | 11 | 3 | 1 tie | 78.6% |
| Against the spread | 6 | 8 | 1 push | 42.9% |
| Game total | 7 | 8 | None | 46.7% |
The Giants–Commanders 27–27 tie is excluded from the outright win/loss count and binary probability scores. Its spread and total predictions still receive their own grades. Miami +7 supplied the spread push when Cincinnati won by exactly seven, 31–24. A push adds neither a correct nor an incorrect spread prediction. Washington’s official schedule, NFL Week 5 results
These are records at the original forecast lines. Later market movement does not replace those benchmarks, and the table measures prediction outcomes rather than returns from a betting portfolio. Spread and total tracking began in Week 4. This Week 5 record adds another week to evaluate; it does not establish a reliable advantage in either market.
Every game, against the saved forecast
The table pairs the separate XGBoost point projections with actual scores, then grades the winner blend and saved spread and total selections. Scores read away–home, matching the listed matchup; LAR means the Rams; Philadelphia–Jacksonville was played in London. W is a correct selection, L an incorrect one, T an outright tie and Push a spread push. These are prediction grades, not wager results.
| Game | Projected score | Final score | Winner pick | Spread pick | Total pick |
|---|---|---|---|---|---|
| TB at DAL | 20.0–28.5 | 24–16 | DAL 59.8% / L | DAL -8.5 / L | Over 48 / L |
| PHI at JAX | 18.3–27.0 | 17–19 | JAX 70.4% / W | PHI +7.5 / W | Over 42 / L |
| CIN at MIA | 23.1–20.9 | 31–24 | CIN 58.6% / W | MIA +7 / Push | Over 42.5 / W |
| LV at NE | 22.8–22.4 | 17–45 | NE 54.3% / W | NE -3.5 / W | Over 45 / W |
| MIN at NO | 20.1–22.3 | 23–13 | MIN 60.9% / W | NO +2 / L | Over 42 / L |
| CLE at NYJ | 21.0–23.7 | 12–17 | NYJ 52.5% / W | CLE +2 / L | Over 40 / L |
| IND at PIT | 21.0–22.4 | 22–34 | PIT 54.1% / W | IND +2.5 / L | Under 44.5 / L |
| HOU at TEN | 19.7–20.0 | 26–3 | HOU 56.2% / W | TEN +7.5 / L | Over 37.5 / L |
| NYG at WAS | 21.0–22.3 | 27–27 | WAS 53.4% / T | NYG +3.5 / W | Over 42.5 / W |
| CHI at GB | 24.8–22.1 | 7–31 | CHI 57.5% / L | CHI -2 / L | Over 46 / L |
| DEN at LAC | 21.5–22.9 | 23–17 | DEN 53.6% / W | LAC +3.5 / L | Over 42 / L |
| DET at ARI | 25.4–27.1 | 45–21 | DET 61.1% / W | ARI +5 / L | Over 54.5 / W |
| SF at SEA | 24.4–23.7 | 32–29 | SF 53.3% / W | SF +3 / W | Over 46 / W |
| BAL at ATL | 22.4–23.4 | 30–23 | BAL 54.3% / W | BAL +3.5 / W | Over 44 / W |
| BUF at LAR | 27.7–28.8 | 37–34 | LAR 50.1% / L | BUF +3 / W | Over 54.5 / W |
The three outright misses were Dallas, Chicago and the Rams. Jacksonville, the strongest saved winner pick at 70.4%, won only 19–17. That also made Philadelphia +7.5 a correct spread selection. Jacksonville remained the favorite; Philadelphia covering did not require an outright win.
Point estimates tell a less flattering story than the winner record. Across all 15 games, the mean absolute error was 12.01 points for the margin, 10.86 for the total and 7.83 per team score. These averages include overtime scoring and the tied game.
New England supplied the largest margin miss: the point model projected Patriots 22.39, Raiders 22.82, but New England won 45–17. That is a 28.43-point margin error, even though the separate winner blend correctly selected New England. Chicago’s 31–7 loss at Green Bay was the next-largest margin miss at 26.77 points. A correct winner selection does not necessarily mean a close score projection.
Buffalo closes the week in overtime
The Rams forced overtime with Kyren Williams’s one-yard touchdown with five seconds left. Buffalo then lost James Cook’s fumble on its opening overtime possession, with Nate Landman forcing it and Aaron Donald recovering. Greg Rousseau kept the Bills alive by blocking Harrison Mevis’s 48-yard field-goal attempt. Tyler Bass subsequently made the winning 45-yard kick. Bills recap, Rams recap
Our frozen win forecast had Los Angeles at 50.1%, essentially splitting the teams. The Rams selection lost, while Buffalo +3 and Over 54.5 both won. The teams combined for 71 points. Buffalo’s spread grade uses the original +3, even though Monday’s market offered +3.5.
The Monday preview projected Rams 28.82, Bills 27.70, a 56.52-point total. Buffalo scored 9.30 more points than projected and Los Angeles 5.18 more, leaving the total 14.48 points too low. The projected Rams margin of 1.12 became a three-point Bills win, a 4.12-point margin error. The saved Buffalo +3 selection covered by six points, and the 71-point final cleared 54.5 by 16.5.
The preview’s watch angles produced concrete moments to revisit. Josh Allen converted a crucial third down with a 15-yard scramble on the fourth-quarter possession that ended with Cook’s go-ahead touchdown. The overtime fumble and blocked kick then showed how quickly a scoring opportunity could change hands. Bills recap
Availability also became clearer before kickoff: DJ Moore, Dee Alford and Landon Jackson were inactive alongside the already ruled-out Joshua Palmer and Ed Oliver. Quentin Lake was not on the Rams’ inactive list. Those facts resolve the preview’s pending designations; they do not quantify an injury adjustment to the saved forecast. Bills inactives, Rams inactives
The other primetime games went in opposite directions for the predictions. Tampa Bay’s 24–16 win at Dallas defeated all three saved picks: Dallas to win, Dallas −8.5 and Over 48. The Friday review details that miss. Baltimore’s 30–23 overtime win at Atlanta made Baltimore to win, Baltimore +3.5 and Over 44 all correct. Cowboys recap, Ravens game center
What the probabilities tell us
The ensemble’s outright log loss was 0.618, with a Brier score of 0.213, across the 14 games that did not finish tied. Both scores evaluate the probabilities assigned to outcomes; lower is better. Elo selected fewer winners, eight, but posted a lower log loss of 0.586. Picking more winners and assigning better probabilities are related measures, but they can disagree.
| Winner forecast | Correct–incorrect | Brier score | Log loss |
|---|---|---|---|
| Elo | 8–6 | 0.1998 | 0.5862 |
| Logistic regression | 10–4 | 0.2282 | 0.6481 |
| XGBoost | 9–5 | 0.2293 | 0.6512 |
| Bayesian | 9–5 | 0.2361 | 0.6630 |
| Random Forest | 9–5 | 0.2057 | 0.5995 |
| Weighted ensemble | 11–3 | 0.2131 | 0.6178 |
Every row also has the same one tied game, excluded from these binary scores. For each decisive game, Brier measures the squared probability error, while log loss penalizes giving the eventual winner too little probability. Elo’s lower scores show why winner accuracy alone is an incomplete model comparison.
The probabilities remain raw and uncalibrated, and one successful week of outright picks does not establish their precision. Spread and total aggregation also remains unvalidated: its weights were inherited from win models, rather than fitted specifically to these markets. The supported spread estimates come from three trained models; Elo and Bayesian spread proxies are excluded. The totals blend also retains neutral 50% contributions, which are not additional trained totals models.
Week 5 therefore leaves a mixed record: the ensemble selected eleven winners, but its spread and total selections each finished below half correct. The separate records and probability scores keep those outcomes visible without treating one successful measure as proof that the rest worked.
Source note: This article uses the local Tuesday, October 13 closeout package based on prediction-repository revision 6083c36, the unchanged pregame forecast and the official sources linked above. Tuesday’s package was not yet committed when this article was prepared; companion editorial evidence preserves its exact source hashes, full-precision projections and settlement calculations. All spread and total grades use the original forecast lines. This is a Week 5 review, not a cumulative dashboard record. Preparing it did not change models, forecasts or grading records.
Written by cresencio
← Back to blog
