NFL Week 4 Recap: Two Props Lost, the 49ers Covered

The receipts stay on the table

My favorite pick lost. My second-favorite pick lost. The one that had me complaining about the injury report won. Tremendous opening demonstration of my ability to rank my own anxiety.

The 2026 NFL Week 4 card finished 1-2-0, down 1.15 units on three one-unit singles. JSN over 7.5 receptions at +105 lost. James Cook over 88.5 rushing yards at -115 lost. San Francisco -2.5 at the recorded 1.85 decimal price won and returned 0.85 units of profit.

That is a minus-38.33% return on the three units risked. An ugly week, not a statistically meaningful verdict on an entire method. These are my independent analyst selections, separate from the official ATSKillers model record. No new picks are being added in this recap.

Here is the part worth discussing: the two props did not lose in the same way as a player simply having a terrible game. The opportunity I needed never arrived. Writing down a risk on Saturday does not give me immunity when it shows up on Sunday. I still chose the bet.

Jaxon Smith-Njigba receptions: six targets cannot become eight catches

Smith-Njigba finished with five catches for 76 yards on six targets. Seattle threw 22 passes and completed 13. He caught five of the six balls sent his way. Calling that a conversion disaster would be bullshit. There were only six chances to convert.

The original argument was that Seattle kept finding him across alignments. That was a legitimate reason to investigate the receptions market. It was not enough, by itself, to justify needing eight catches. A strong role can survive while the team's passing volume shrinks underneath it.

Seattle led 27-6 in the third quarter and finished with 33 rushing attempts. The Seahawks' own postgame account reports 138 rushing yards and four takeaways. They had ways to win that did not involve feeding my favorite receiver eight completions. Very inconsiderate of them to prioritize their actual job.

JSN received about 27% of Seattle's pass attempts and caught about 83% of his targets. Those are observed, rounded ratios, not stable rates to paste onto next week. His role did not disappear. The available passing pie was small, and a healthy slice still left the receptions over short.

My process criticism is that I gave the recent catch totals too much persuasive weight without making the lower-volume path concrete enough. I mentioned the run-heavy lead in the preview. I did not adequately establish why the over remained attractive when that happened. Mentioning the trap and walking into it is still walking into it.

The better question before the next receptions pick is how the team gets to enough attempts, how many of those become this player's targets, and what happens if either count falls. I have not verified route-by-route participation or coverage assignments for this review, so I am not inventing a defensive scheme explanation.

James Cook rushing yards: the running was fine; the twelve carries were not

Cook ran for 75 yards on 12 carries. Divide the actual totals and you get 6.25 yards per carry. That is good production per attempt. It still loses an over 88.5 ticket because the sportsbook does not pay extra for doing your work efficiently.

Buffalo ran the ball 20 times as a team. Cook had 12 of those attempts, Josh Allen had five, and Ray Davis had one. So I am not blaming this on Davis suddenly taking a giant workload. Cook's share was 60% of team rushing attempts, and the entire rushing budget was only 20.

The Bills lost 29-26. Their official recap connects Cook's limited carries with time spent behind, although Buffalo did lead 17-14 at halftime. This was not a game in which they trailed by four touchdowns from the opening whistle. Game state matters without becoming an excuse that explains every play call.

My preview used 20 carries at 4.5 yards per carry as an illustration: 90 yards, just enough. That was a scenario, not a saved statistical forecast. It exposed the dependency but did not establish how likely 20 carries were. I needed to press harder on that distinction before choosing the over.

Here is the arithmetic of the miss against that illustration. Moving from 20 carries to 12 at the reference 4.5 rate removes 36 yards. The actual efficiency adds 21 yards back across those 12 carries. Ninety minus 36 plus 21 equals 75. This is bookkeeping, not proof of what would have happened on eight imaginary extra runs.

The line needed at least 89 rushing yards. On 12 carries, that would require about 7.42 yards per attempt. Cook was productive and still would have needed substantially more efficiency to rescue the bet. The weakness was relying on enough work, not discovering afterward that he had forgotten how to run.

I will keep separating team rushing volume from the player's share. A favorable run-blocking or player-value profile helps describe what someone might do with a carry. It does not compel the offense to hand him the damn ball.

49ers -2.5: take the win, leave the victory parade

San Francisco won 24-14 and covered -2.5. Purdy finished 19-of-30 for 218 yards and two touchdowns, with no sacks. That protection is consistent with a part of the case I liked. The passing offense contributed. The ticket won. We can say both without pretending the whole afternoon was a clinic.

The official recap also records a third-quarter special-teams takeaway that gave San Francisco possession at Denver's 28. That possession became the touchdown for a 10-3 lead. Later, with the margin down to three, Kittle's 56-yard touchdown stretched it to ten. The final margin contains a short field and a big late play, not only a steady march of ordinary offensive superiority.

That is why the winning pick gets reviewed too. If I explain losses through every ugly detail and wins through my own brilliance, you should confiscate the keyboard.

The exact entry was still -2.5 at 1.85 decimal. A three-point win would have mattered to that number, but the actual ten-point win also covered -3.5. So no, the half point did not save this particular result. It was good entry discipline, not the reason this Sunday paid.

I am retaining the original positive passing-game assessment as a hypothesis that had some support, while acknowledging the assistance and the narrow late-game margin before Kittle scored. One cover does not validate an injury adjustment I never measured or turn a research index into a fair spread.

What changes in my preparation

I want to use more of the existing research properly, not announce a miracle model because I had a bad Sunday. Rolling player values, team-unit values and retained trends can tell me where to look. They do not, on their own, tell me how many opportunities a player will receive or the probability of beating a particular price.

My private analysis now separates the quality evidence from an opportunity worksheet. For a receiving scenario: team pass attempts, target share, catch rate, then yards per catch if that market is relevant. For rushing: team attempts, player share, then efficiency. Player allocations cannot add up to more than the team's available work.

I want a lower-opportunity case on the page before making the pick, not buried in the last sentence. A player can remain excellent in that case and still be a bad choice at the offered line. I also want to know which assumption actually needs to change for my opinion to change.

Those calculations are conditional scenarios, not calibrated hit rates. I have not trained a new predictive system or demonstrated that it beats the market. Three selections cannot establish that. Any future numerical betting-probability model would need separate, chronological testing against untouched outcomes and actual prices.

Trends need dates and definitions, too. A rating change measured from an old season endpoint is not automatically recent form. A data-confidence score is not confidence that a bet wins. Reading the label is less glamorous than drawing a giant green arrow, but it keeps me from publishing nonsense with excellent typography.

For this recap I compared the original pregame case with independently checked final statistics. I did not rerun the old picks using newly updated player values and call that foresight. The new worksheet is for future questions; it does not rewrite the receipt. The official ATSKillers model and its records remain separate and unchanged.

What this review cannot tell us

I do not have a verified closing-price comparison for all three saved selections, so there is no closing-line-value victory speech. The original card was frozen Saturday without a complete final-inactives and weather review. That limitation remains in the history; this recap does not invent the missing checks.

Nor does the final box score isolate how much every injury, coverage adjustment or coaching choice caused the result. It tells us enough to reject a few lazy explanations. JSN did not miss because he dropped six targets. Cook did not miss because 6.25 yards per carry was inefficient. San Francisco did not cover solely because my opinion existed.

The useful lesson is to demand better evidence for the opportunity a pick requires, while keeping the uncertainty visible. It is not to blindly bet unders next week, increase the stake to get even, or declare every winner a good decision.

TL;DR: two losses, one win, no disappearing ink

JSN: five catches on six targets. The eight-catch bet ran out of opportunities. Cook: 75 yards on 12 carries. Strong efficiency could not repair the workload shortfall. San Francisco: a 24-14 win, with sound protection, a useful short field and a big late touchdown.

The card stays 1-2-0 and minus 1.15 units. The research gets a more explicit opportunity check. The stakes do not go up because I am annoyed, and the losses do not get renamed leans.

Keep the original article next to this one. That is the whole point of putting the reasoning in public: you get to check whether Monday's explanation resembles what I actually said before kickoff.

Market references & reading

• Original pregame article
• Independent Nori record catalog