New England Patriots 29, Buffalo Bills 26

Opening and closing line

DraftKings via ESPN, one book. Not the Kalshi price — that block reads our own tape.

Spread

Bills −3 → Bills −7

Total

49.5 → 49.5

Moneyline

NE +136 / BUF −162 → NE +275 / BUF −345

The spread moved toward the side that lost.

How it happened

What this recap does not know

• —The ledger has not graded this game. 2026_04_NE_BUF_moneyline and 2026_04_NE_BUF_total carried no result, pnl_cents or graded_at when this was written (2026-10-04, 22:10 UTC); the outcomes above are derived from the final score, and the prop grades from the ESPN box score by script (handoffs/artifacts/NFL_W4_SUNDAY_EARLY_2026/prop-reconciliation-w04-sunday-early.csv), never by hand. entry_price on a NO or UNDER row is the YES price, so that side’s cost is 100 minus it, and no P/L is summed on this page. If the ledger ever disagrees with anything above, the ledger is right and this page is wrong.
• —The Patriots moneyline row was captured 2026-09-22, twelve days before kickoff. The preview on this page carries no game-line call, and Saturday’s article called the under the one game position on its card. Whether the Patriots row showed on the edges board on Sunday morning was not checked against a page snapshot.
• —Cook’s 2+ receptions cleared on exactly two catches. A stat correction taking one away would flip that grade.
• —DJ Moore (shoulder, did not return), Ed Oliver (knee), Dee Alford (ankle) and Mack Hollins (calf) had no confirmed severity when this was published.
• —ESPN’s summary carried no weather block, so the wind input behind the total (Saturday’s article: about 14 mph in the engine, 6 to 7 mph on the National Weather Service point forecast) was not checked against observed conditions. Attendance (60,606) is ESPN’s figure.
• —This recap does not know what the Kalshi book did during the game. Nothing above is a live price; the entries and closes quoted are the ledger’s captures, the DraftKings numbers are ESPN’s archived open and close, and the prop grade is a resolution against the box score, not a settlement read off the exchange.

What this read does not know

• —The injury feed we read is practice participation, not game-status designations. Our Week 4 report is practice participation only and was fetched 2026-10-01 12:01Z, so nothing either club files on Friday is in it. Every body named above is Wednesday-accurate and no more — a DNP on Wednesday is not an OUT, and New England’s three absences are the kind that get upgraded by Friday as often as not.
• —The power ratings behind the model line are this week’s board (as-of 2026-09-30, written by the Week 4 pipeline run, with Buffalo 4th at 3.2). An earlier version of this read called them stale; that was a freshness label on our own tool scoring a weekly board against a daily clock, fixed on 2026-10-01.
• —Josh Allen may have tweaked a knee and is not on the injury report. Two independent previews note he appeared to hurt his left knee late in the fourth quarter against the Chargers and ran only one designed run afterwards. He is absent from the Wednesday report entirely, which is strong evidence and not an explicit full-participation line. Allen’s legs are a large share of what makes Buffalo’s offense what it is, so a quiet knee is the single biggest thing we cannot price here.
• —T.J. Sanders’s absence is reported two different ways. The clubs’ Wednesday filing lists him DNP with an illness; a secondary outlet reports him out following an appendectomy, which is a far longer timeline. Where the two conflicted we used the club filing’s wording, but we do not know which one describes his Sunday.
• —Christian Benford’s toe is the hinge and we have no read on it. He left Sunday’s game late in the third quarter, wore a boot on his right foot afterwards and did not practise Wednesday. Buffalo has not put a timeline on it. The whole New England passing path in the third read above weakens if he plays and strengthens if he does not, and we will not know which until after this module’s pull window.
• —We do not know which New England defense shows up, and the Buffalo turnovers may be noise. Buffalo gave the ball away five times and still won in Week 3, and its offensive line played badly in that game — so the one way the Patriots stay in this is a repeat of Buffalo’s worst sixty minutes, not anything New England has done on its own. Three games is not enough to tell a 3-0 team’s floor from its variance, and we are reading both of those off the same small sample.
• —James Cook under 89.5 rushing yards is a WATCH, and the fork is his workload. We project 78.3 yards and our model has him over 89.5 about 37% of the time — but that projection carries him at roughly 14 carries, and he has had 13, 21 and 24, running for 135 and 154 on the two big ones. If Buffalo keeps him at 20 or more, the under is fighting his last two games and we would not take it. If the Week 1 split comes back (13 carries, 57 yards), it is live. Which Buffalo shows up is the whole question, and the model cannot see it.