Buccaneers beat Cowboys as three Week 5 picks miss

Tampa Bay finally has its first win, and our Week 5 predictions have three opening misses. The Buccaneers beat Dallas 24–16 at AT&T Stadium on Thursday, October 8, moving to 1–4 while the Cowboys fell to 2–3. Our saved picks were Dallas to win, Dallas to cover an 8.5-point spread, and the game to finish over 48 points. All three were wrong. Cowboys game recap

Dallas led 10–7 at halftime, but Tampa Bay changed the game on the ground after the break. Bucky Irving’s 72-yard run on the opening possession of the second half set up Emeka Egbuka’s 14-yard rushing touchdown. The Buccaneers eventually built a 24–10 lead, and Alex Anzalone’s late interception helped close out the win. Tampa Bay finished with 241 rushing yards and a 3–0 advantage in takeaways. Cowboys game recap

Irving supplied 165 rushing yards on 21 carries and scored once on the ground and once as a receiver. Jalon Daniels completed 19 of 25 passes for 189 yards, one touchdown and no interceptions. That combination gave Tampa Bay enough offense to turn its halftime deficit into a two-score lead. Official box score

How the saved predictions graded

Here is how the original predictions graded against the final score. These are the probabilities saved before kickoff; the result has not changed them.

The Week 5 record therefore opens at 0–1 in each category. Dallas’s cover calculation was 16 − 24 − 8.5 = −16.5; Tampa Bay +8.5 covered comfortably and also won outright. The 40-point total finished eight below the saved line. These are three related misses from one game, rather than three independent tests.

Spread and total grading uses the original −8.5 and 48 benchmarks, not a verified closing line. The forecast was generated at 3:17 a.m. Central on Thursday and issued at 3:39:58 a.m., well before the game. The Thursday preview preserves the predictions and their context.

The point model overestimated Dallas’s offense

The preview also recorded separate XGBoost margin and total projections, with team scores derived from (total ± margin) / 2. Here is how those estimates compare with the final score:

Dallas scored 12.47 fewer points than projected, while Tampa Bay scored 3.99 more. The margin missed by 16.46 points, and the total was 8.48 points too high. Those errors describe the forecast’s performance; they do not identify the cause of the miss.

The unrounded Dallas margin was 8.4567, just below the 8.5-point handicap. That standalone projection did not clear Dallas −8.5 even though the separate spread classifier favored a Dallas cover. Both were far from the actual result.

Every winner model missed, with different penalties

Our Dallas win estimate was less decisive than the five-model agreement might suggest. Every base model selected the Cowboys, but their Dallas probabilities ranged from 50.1% to 68.6%. XGBoost was almost evenly split. The ensemble assigned Tampa Bay a 40.2% chance of winning, but Dallas was still its selected winner and takes the loss in the record. Agreement on a team is different from agreement on how likely that team is to win.

Friday’s report grades every winner forecast using Brier score and natural-log loss; lower is better. With Dallas losing, the calculations are p² and −ln(1 − p), where p is its full-precision pregame win probability.

XGBoost receives the smallest penalty because it was least confident in Dallas. Logistic receives the largest. Neither conclusion changes the winner record: all five models and the ensemble opened Week 5 at 0–1.

For Thursday’s Tampa Bay–Dallas matchup, the last saved BetMGM prices were Dallas −475 and Tampa Bay +360, observed Wednesday at 8:59 p.m. Central. Removing the quoted margin produced roughly a 79.2% Dallas probability, compared with the model’s 59.8%. These are historical prices for the completed game, not verified closing odds.

The probabilities remain raw and uncalibrated. Thursday adds three incorrect picks to the record; it does not establish why the models missed or resolve how well their probabilities track actual outcomes. The spread and total estimates also belong only to their original lines.

Friday market update

Our Friday 7:22 p.m. CDT BetMGM snapshot covers the 14 remaining games. Five spreads and ten totals changed from Wednesday evening. The largest total moves were Cincinnati–Miami rising from 42.5 to 43.5, Indianapolis–Pittsburgh falling from 44.5 to 43.5, and Giants–Washington falling from 42.5 to 41.5. All five spread changes were half a point.

Jacksonville remained a 7.5-point favorite against Philadelphia, with the total down from 42 to 41.5. Detroit moved from −5 to −5.5 at Arizona, with the total holding at 54.5. Minnesota moved from −2 to −2.5 at New Orleans, while that total fell from 42 to 41.5. These are Friday observations; the saved probabilities remain unchanged and do not supply new spread or total estimates at the moved lines.

Sunday outlook

Fourteen Week 5 games remain: thirteen Sunday and one Monday. Sunday starts with Philadelphia against Jacksonville at Tottenham Hotspur Stadium at 8:30 a.m. Central on NFL Network. Jaguars viewing guide

Jacksonville leads the saved Sunday forecast at 70.4%, with four of five models choosing the Jaguars and Elo leaning Philadelphia. Detroit follows at 61.1% at Arizona, with Minnesota at 60.9% at New Orleans. Those remain the original win estimates as the rest of Week 5 approaches. The full Week 5 preview keeps the complete forecast board available for comparison.

Source note: This review uses the refreshed Friday draft and operations report committed at prediction-repository revision e668e21, the unchanged pregame forecast, and the official game sources linked above. The companion editorial evidence preserves source hashes, full-precision forecasts, settlement calculations and Friday’s market snapshot. Preparing this article did not refit models or change forecast or grading records.