NFL Week 7 Preview: Houston Leads, Primetime Picks Need a Closer Look

Houston leads the Week 7 winner picks at 73.7%, followed by Minnesota at 67.9%, Carolina at 65.6% and San Francisco at 64.3%. All five base models choose each of those teams. The primetime games are less straightforward: Chicago has a moderate Thursday lean, Kansas City–Seattle divides the models, and Dallas–Philadelphia exposes a disagreement between the separate winner and spread probabilities.

This preview covers all 14 games, October 22–26. The forecast was issued Tuesday, October 20, at 3:07 p.m. CDT and remains frozen. Wednesday’s market update shows what changed in the available prices; it does not rewrite the picks.

Week 6 finished 7–7 on winners, 4–9–1 against the spread and 9–5 on totals, graded at the original lines. Its three primetime games went 3–0 on winners and spreads, but the full slate was much less successful. That is useful context before reading this week’s strongest probabilities as guarantees.

Green and a checkmark identify the predicted outright winner. Spread and total selections are separate and appear in the tables below.

The three primetime games

Thursday, October 22: New England at Chicago, 7:15 p.m. CDT. Chicago is a 57.7% winner pick, with all five models favoring the Bears. Their estimates range from 53.3% to 61.9%, so agreement means a shared lean rather than overwhelming confidence. The original selections are Chicago −3 at 54.0% and Over 50.5 at 52.2%. Patriots game preview.

New England’s protection and ball security are worth watching. The Patriots’ Wednesday matchup analysis reports a minus-six turnover margin through six games, while Caleb Williams completed 30 of 36 passes in his return last week. Chicago forcing short fields would make the three-point handicap harder for New England to manage. The available injury report still lists Tuesday participation, with final game designations pending; a missed practice is not a declaration that a player will miss Thursday. Patriots matchup analysis, official injury report.

Sunday, October 25: Kansas City at Seattle, 7:20 p.m. CDT. The ensemble picks Kansas City at 56.9%, but only three of five models choose the Chiefs. Individual Seattle win estimates range from 20.3% to 60.5%. The original spread lean is Kansas City +2.5 at 55.5%, with Over 42.5 at 52.8%. Wednesday’s BetMGM quote still makes Seattle the market favorite at −145, with Kansas City +120: the model and market prefer different winners.

Seattle’s protection is a clear watch point after allowing five sacks and 17 quarterback hits against Denver. Sustaining drives at Lumen Field would give the Seahawks a better chance to justify that favorite status. Those previous-game numbers provide football context; they do not explain the price movement or change the saved forecast. Seahawks coach update, official game information.

Monday, October 26: Dallas at Philadelphia, 7:15 p.m. CDT. Dallas is the 55.5% winner pick, with four models on the Cowboys. Elo is the outlier, assigning Philadelphia 70.4% to win. The original total lean is Under 44.5 at 52.5%, while Wednesday’s saved BetMGM total has moved to 44. The old probability belongs to 44.5. Eagles schedule.

The spread output needs an explicit caveat: Dallas −4.5 is assigned 57.6% to cover, above its 55.5% chance of winning outright. Covering −4.5 requires winning, so these separately fitted models do not form a consistent set of probabilities. Both numbers remain visible for transparency; they should not be described as reinforcing confidence in Dallas.

Philadelphia’s quarterback situation is another reason to revisit the matchup before kickoff. The team’s latest game analysis examines an offense led by Tanner McKee while Jalen Hurts is on injured reserve. Sustaining drives with that offense will be a central question against Dallas. The Monday post will need the latest team update; this preview does not assume a manual adjustment to the frozen predictions. Eagles offensive analysis.

Where the Sunday board gets interesting

Pittsburgh–New Orleans opens Sunday at 8:30 a.m. CDT in Saint-Denis, France. New Orleans is the designated home team at Stade de France. Pittsburgh is 59.8% to win, while the separate spread model slightly prefers New Orleans +3.5 at 51.4%. A Steelers win and Saints cover can both happen. Saints schedule.

Houston’s leading win probability does not promise a large margin. The Texans are 73.7% to beat the Giants, but only 52.1% to cover the original −9.5. Wednesday’s BetMGM handicap is now −10, where that probability cannot be reused. Under 43 is almost neutral at 50.1%.

Minnesota and Carolina have different spread questions. Minnesota’s 67.9% win estimate comes with 52.0% to cover −6.5. Carolina is 65.6% outright and 55.5% at the original −3, but the Wednesday quote has moved to −3.5. Losing the possibility of a push on a three-point win matters even when the selected team stays the same.

Baltimore is the narrowest winner pick at 54.0%. Only three models choose the Ravens, and Wednesday’s market moves from Baltimore −3 to −1. Its original Over 51 at 54.3% and San Francisco–Atlanta’s Over 50.5 at 54.2% are the largest saved over leans. Both totals have moved lower, so neither old probability describes the new line.

Detroit–Green Bay and Rams–Raiders offer different kinds of disagreement. Detroit is 60.5% to win, while Green Bay +3 is the slight spread choice at 51.7%. The Rams are 59.3% outright with a stronger 56.9% original −4.5 cover lean, yet the separate point model projects a narrow Las Vegas win. Model outputs need to be read individually.

What the point model projects

These scores come from the separate XGBoost margin and total models, not the winner-probability blend. Team scores and totals are rounded independently from the full-precision projections.

Chicago’s projected 2.33-point margin is smaller than the original three-point handicap. Pittsburgh’s margin is just 0.13 points, despite its 59.8% winner estimate. Dallas’s projected 0.56-point margin is also much smaller than 4.5. None of those margins is itself a cover probability.

The total projections can disagree with the separate total selections, too: Pittsburgh–New Orleans projects 46.01 points, while the total blend slightly prefers Under 44.5. Keeping both visible makes the limits of these separate models easier to see.

Wednesday’s market update

The new snapshot was collected Wednesday, October 21, at 1:02 p.m. CDT, compared with Tuesday, October 20, at 12:55 p.m. CDT. These are saved observations, not live or verified closing prices.

The comparison uses BetMGM throughout. Moneylines and spreads are for the home team, including designated-home New Orleans. Parentheses show the American price for that spread. Totals are the game lines. All arrows compare Tuesday with Wednesday at the same bookmaker.

Six spreads and five totals differ from the frozen forecast lines. Baltimore −3 to −1 is the largest spread move in this table. Atlanta +3.5 to +3 and Carolina −3 to −3.5 move onto or off the three-point number. These observations establish the changes, not their causes.

Thursday is a price-change story: Chicago stays −3, but its price improves from −108 to −102. New England +3 becomes more expensive, moving −110 to −118. The total remains 50.5. A stable handicap does not mean an unchanged offer.

The separate Wednesday screen found 41 qualifying offers across bookmakers: 31 moneyline and 10 spread offers, consolidated into eight recommendations: five moneyline and three spread selections. No total qualified. Multiple books can quote the same selection, so offers and recommendations are different counts. These are saved screening results, not placed bets or realized returns. Probabilities were not reassigned to changed lines.

All fourteen frozen picks

All kickoff times are CDT. “Models” counts the five base models choosing the ensemble’s outright winner; they are not independent votes. LAR means the Rams. Pittsburgh–New Orleans is at a neutral venue, with New Orleans designated home. Spread and total probabilities below apply to their original lines, even when Wednesday’s quote differs.

*Dallas’s cover probability exceeds its win probability because those outputs come from separately fitted models. This is a consistency problem, not extra evidence for the pick.

How much weight to put on the numbers

Monitoring is GREEN, with winner log loss of 0.6802, averaging the three weekly scores equally across the original Weeks 4–6 results. A separate two-week winner-accuracy window, Weeks 5–6, is 18 of 28, or 64.3%. Those are different windows. GREEN means the configured warning thresholds are not currently breached; it does not validate calibration or demonstrate profitability.

The forecasts still use raw, uncalibrated probabilities. A 60% estimate is not an established six-in-ten hit rate. Verified recommendation ROI remains unavailable because the monitoring inputs contain no usable settled bets.

The spread blend uses three trained market models. The totals blend retains two constant-50% components alongside three trained models, and integer totals have no separately saved push probability. These market models do not have the same weekly validation coverage as the winner ensemble. That is why an outright pick, a projected score, a cover estimate and a qualifying priced offer should not be treated as interchangeable.

Thursday’s final designations and inactive lists, later injury reports and game-day conditions belong in the matchup updates. The original forecast stays fixed so the eventual results can be judged against what was available before kickoff.

Source note: Forecasts and market observations come from prediction-repository revision 03c1bdf. The Week 7 forecast was issued October 20 at 20:07:52 UTC; the Wednesday capture was collected October 21 at 18:02:06 UTC. Companion editorial evidence retains the exact probabilities, original lines, source timestamps and file hashes. Official team sources support the football context. No models were retrained or forecasts regenerated while preparing this post.