Buffalo Bills at Houston Texans. Lean: Houston Texans +1.5. Lean: Under 44.5.

Spread.Take Houston +1.5, listed at -118 by DraftKings. Our Houston +0.4 makes this only a 1.1-point disagreement, insufficient evidence by itself given the model’s poor ATS record. The better argument is Houston’s pressure matchup: its official September 8 preview documents Allen’s eight sacks in their latest meeting. That supports a competitive home underdog, although repeating that extreme outcome would be an unreasonable assumption. Buffalo has DJ Moore, and Shakir’s September 9 practice appearance reduces concern about its receiving options. Houston’s Braden Smith absence and rookie interior starter subtract from the Texans case. DraftKings shows 76% of spread tickets and 73% of handle on Buffalo, but the dossier shows its spread unchanged since August 30; that is resistance, not proof of professional Houston money. Allen’s advantage and Buffalo’s offensive continuity under promoted coach Joe Brady are already plausible reasons for road favoritism. I narrowly prefer Houston’s defense plus the points, with unresolved practice participation limiting conviction.

Total.Take under 44.5, available at -105 on the consulted DraftKings page. The serious over case is the bias-adjusted model’s 47.95, Allen’s scoring ceiling, Moore’s addition, Shakir returning to practice and likely indoor conditions. Nevertheless, Houston’s pressure can end Buffalo possessions, while Smith’s absence and Rutledge’s debut introduce Houston protection risk. I place more weight on those matchup constraints than the model’s unexplained Houston offense bump. DraftKings’ 75% under handle against 49% of tickets is supportive context, not proof of informed action. No verified 2026 pace evidence justifies a tempo adjustment. Because the market already prices defensive strength and the model points the other way, this is only a marginal under preference.

Where our number looks wrong.Frozen ratings miss Smith’s subsequent injured-reserve placement and Shakir’s current uncertainty. A new-head-coach flag also cannot distinguish Brady’s offensive continuity from Buffalo’s defensive transition under Jim Leonhard. The totals projection looks too optimistic about Houston: subtracting the stated 1.2-point bias reduces 49.15 to 47.95, but its additional 1.995-point Houston offense bump remains difficult to justify against offensive-line uncertainty. Roof closure is an assumption, not verified news.

Spread.Our v3 rating makes this Buffalo by 0.4; the market is Buffalo -1.5. That 0.8-point disagreement leans Houston, but the model is 47.6% ATS on held-out data, so it is a tiebreaker, not a signal. The news breaks the tie. Houston's defense was the NFL's top unit in 2025 at 277.2 yards allowed per game (NBC Sports, 9/9/26) and added Reed Blankenship, Logan Hall and Kayden McDonald (buffalobills.com, 9/9/26). Buffalo ranked 28th against the run last season and debuts a brand-new Jim Leonhard scheme in Week 1 (buffalobills.com, 9/9/26) against a Montgomery/Marks committee. Allen has never won in Houston: five straight Bills losses there, a 64.2 rating in three road games (buffalobills.com, 9/9/26). Buffalo's skill group is patched, with Shakir practicing Wednesday for the first time since his injury and Coleman off a foot sprain. Houston's own losses (Higgins, Dell, Braden Smith) are real and are the reason this is priced at a pick rather than a Texans number, but at +1.5 at home I take the points.

Total.Our totals model says 49.2 against a market of 44.5-45, but it is documented as worse than the market (MAE 10.46 vs 10.19), runs about 1.2 points high, and it applied a dome bump on an imputed closed roof rather than a confirmed one. Strip the bias and the imputation and most of the over lean disappears. Houston went under in 11 of 19 games in 2025 (NBC Sports, 9/9/26) and now opens without Higgins, Dell and RT Braden Smith, with a rebuilt line in front of a Stroud who threw seven playoff interceptions. Buffalo's 28th-ranked run defense invites a Montgomery/Marks clock-burning plan, not a shootout, and Allen faces the league's No. 1 defense by yards. The number opened 45.5 in May and has been bid to 44.5.

Where our number looks wrong.Three specific holes here. (1) The ratings are frozen at 8/19 preseason, so they cannot see the two things that actually define this game: Houston losing Jayden Higgins for the season, Tank Dell for at least four games and RT Braden Smith to IR, and Buffalo adding DJ Moore. The model is rating a Houston offense that no longer exists. (2) The injuries table is empty, so every one of those facts had to come from outside the model. (3) The new-head-coach flag is set for Buffalo and implicitly treated as disruption, but Joe Brady was already the play caller since 2023; the real Week 1 uncertainty in Buffalo is on defense under new DC Jim Leonhard, which the model does not encode at all. The model's 0.8-point lean toward Houston happens to point the same way I do, but it gets there for none of these reasons, and with a 47.6% held-out ATS record I would not have followed it if the news had pointed the other way.

Spread.So the number itself is a pass; the lean here is qualitative. Houston's starting RT Braden Smith is on IR (out at least 4 games, Yahoo Sports 9/9), a real hit to pass protection for a QB (Stroud) already criticized for his playoff performance. Buffalo's coaching change is lower-risk than a typical new-HC discount: Joe Brady is an internal promotion from OC, keeping offensive continuity around Josh Allen. Buffalo did lose defensive pieces (Milano, Taron Johnson per SI 9/9) and has several questionable/out complementary players (Shakir Q, Shavers out), but its offensive identity is intact. I lean Bills -1.5, with only moderate conviction given both rosters are banged up and no model edge supports either side.

Total.Our totals model reads 4.4 points over the market (49.15 vs 44.7), but its own holdout already runs about 1.2 pts hot on average, so the real signal is closer to a 3-pt over-lean, still large. I fade most of that gap: Houston lost two receivers to injury (Jayden Higgins, season-ending torn ACL; Tank Dell, IR, out 4+ games, Yahoo Sports 9/9), thinning Stroud's options, and Buffalo is breaking in first-time HC Joe Brady for a Week 1 road opener, which historically trends toward early-season conservatism and rust rather than shootouts. NRG Stadium's roof will likely close given a forecast near 93°F with a 50% storm chance (NWS, 9/9), which is close to neutral for pace and passing efficiency either way. I take the under at 44.5, a moderate lean given our own model disagrees with me sharply.

Where our number looks wrong.The model is frozen preseason and its injuries table holds zero NFL rows, so it cannot see Braden Smith's IR placement or weigh Houston's thinned-out receiver room. It also has no mechanism to distinguish an internal OC-to-HC promotion (Brady, lower continuity risk) from a true outside coaching hire, even though it correctly flags Buffalo as a new-head-coach team. With only 47.6% ATS on holdout, even a real disagreement with the market here would not be independently actionable, the market's -1 to -1.5 should be treated as the better-informed number, and my lean is built mostly on the roster news above, not the model gap.