He has beaten his expected points in 2 of 4 seasons. Whether that is a skill or a countdown decides what he is worth at 11.9 overall.
James Cook scored 323.4 fantasy points in 2025. The volume he earned was worth 292.2. That gap of 31.2 points is the single most important thing to understand before drafting him at 11.9 overall.
Rather than reading a season average, it is worth finding the point the season actually turned. Our shift detector flags in-season changes large enough to clear a statistical threshold, then asks what moved around them.
The largest single move was Week 10: stacked-box rate went up from 21.9 to 40.9, a 86.8% change, at 3.37 standard deviations. The cause was structural — the play-calling changed. He also missed Week 7.
Year over year the engine classifies 2025 as a stable year. That is the label to carry into the projection, because it separates a player who lost snaps from one who kept them and converted worse.
He finished 17.02 points per game on 46 touches — down 0.3% from his 2024 rate with BUF. Every part of that is real and none of it is in dispute.
What is worth separating is how much of it he earned. Expected points strip out the finishing and ask a narrower question: given the targets he drew, where they came from on the field, and how far downfield they travelled, what was that usage worth? For 2025 the answer was 292.2 points — 15.4 per game. He scored 31.2 more than that, a 10.7% overshoot.
The expected line underneath the season was 36 catches for 271 yards and 1.7 touchdowns, and he cleared it.
The expected line was 36 catches for 271 yards and 1.7 touchdowns. He finished with 37, 320 and 2.
64% of the entire gain came from receiving yards alone. He hit 103% of his expected catches and 118% of his expected touchdowns — and 118% of his expected receiving yards.
4 of his 17 games produced 25 or more points: Week 2 against NYJ (26.5), Week 8 against CAR (33.6), Week 15 against NE (31.1), Week 16 against CLE (26.4). Strip them and the remaining 13 games average 14.2.
Spike weeks win fantasy games and a player who produces them is more valuable than one who never does — he is a ceiling play with real weeks off. 4 games came in under 10 points, including Week 5 against NE (4.9) and Week 6 against ATL (8.7).
The three largest overshoots — Week 8 (+21.6), Week 16 (+13.1), Week 15 (+12.5) — are the same games. The spikes were not extra volume. They were finishing.
1 of his 4 seasons ran hot: 2025 (+10.7%). Across all 4 he has scored 123.4 points more than his usage was worth.
The reflexive read is regression. Finishing above expectation is unstable, and a player who does it repeatedly is usually one whose luck has not run out yet.
But an expected-points model has a specific blind spot, and it is worth naming: it converts usage at league-average rates. A target twenty yards downfield is worth what the average receiver does with a target twenty yards downfield. If a player is reliably better than average at winning a particular kind of target, he will beat the model every single year, and calling that luck is just mislabelling a skill.
So the question is not whether he beat the number. It is whether there is a mechanism.
Across 185 targets and four seasons, he has been a fundamentally different receiver depending on what the defence played. Against man he has caught 73.2% for 8.7 yards per target and 0.39 expected points added per target. Against zone: 84.2%, 5.92 yards, 0.115 EPA.
That is a gap of 0.275 EPA per target on a sample of 71 man targets against 114 zone. The catch rate goes the other way — he catches more against zone — which is exactly what you would expect from a receiver whose value is in winning contested downfield throws rather than in finding soft spots underneath.
The shell detail sharpens it. Cover 1, single-high man is the look he has faced most: 54 targets, 8.83 yards per target, 0.223 EPA. His best is Cover 3 at 0.251 on 32 targets.
The weakness is specific. Against Cover 2 he has caught 83.7% for 5.6 yards a target and -0.001 EPA across 43 looks — a gap of 0.3 expected points a target between his best coverage and his worst.
This does not prove the overshoot repeats. It does mean the pattern has a stated cause rather than being a coin that keeps landing heads.
Across 120 graded prop lines, he has gone over 49.2% of the time. On rushing yards specifically the number is 60.0% across 30 lines, beating the posted number by an average of 22.6%.
A projection model can be wrong about conversion. A book adjusting weekly, with money on the other side, being wrong 60% of the time on the same market is a harder thing to explain away.
Touchdowns are the control. His anytime-TD market implied a 54.3% rate and he converted at 67.7% — a gap of 13.4 points, which is worth noting.
Buffalo returns 3 of 6 continuity slots — qb1, rb1, te1 — with Josh Allen at quarterback, Joe Brady as head coach and Pete Carmichael Jr. calling it. The offence projects to 31.46 pass attempts a game at medium confidence.
The pie is stable. His slice is the question. Our share model has him at 9.4% against the 10% he ran, with a range of 8.5% to 10.3%.
That number is built from 25 comparable players — same age band, same prior share, same team, same coordinator — not from a depth chart. The closest three went Joe Mixon 10.8% to 13.4%, Jonathan Taylor 10% to 10.1%, Kyren Williams 10.5% to 8.4%. Two held, one fell hard.
Across 2,022 player-seasons we swept every stat for whether it predicts NEXT year's points, and kept only the ones that replicated in all five. f_timeToLOS (r 0.128), f_snapTrend (r 0.103), f_fantasyForm (r 0.078) carry. week-to-week volatility (r -0.163) and receiving EPA per game (r -0.143) run negative — a player who was hyper-efficient on low volume tends to come back down, so a gaudy per-touch number is a warning rather than a selling point.
So the figures worth weighing for him: a 9.7% target share, 0.7% of the air yards, a WOPR of 0.15, on the field for 57.2% of snaps. He finished 5 at the position — RB1.
Vegas sets an implied total for every team every week, and some offences beat theirs systematically. Over 114 games, BUF has come in +2.52 points against its own implied total, clearing it in 61% of weeks.
That is a real structural tailwind. Every skill player in this offence gets a small, repeatable bonus the market prices late.
One related finding worth carrying: across 1,280 cases we measured what happens after a team's implied total slides, and the correlation is -0.37 — negative, meaning the market over-corrects. A falling team total is more often a buying window than a warning.
Our board has him at 266.9 points, 16.7 per game across 16 expected games — 33 catches, 277 yards, 2.5 touchdowns. That is RB1, 12 overall, and 112 points of value over a replacement starter.
It is a step down from 2025, and the reason is volume rather than efficiency. The projection assumes he keeps doing what he does on slightly fewer looks.
zero systemic risk (Buffalo's going to score, he's the clear early-down/TD guy), but the swing factor has always been receiving work specifically — Cook's production splits hard by game state (much worse in losses than wins, unlike Gibbs/Bijan/CMC who don't have that swing). Brutal opening-month schedule (Texans/Lions/Chargers/Patriots/Rams) flagged as a real near-term drag.
That is the current read. It is a change — in 2026-06-05 the position was HESITANT: Bills have the hardest opening schedule and Cook's receiving cratered in past losses, so a slow start is likely. Don't pay the tail-1st price in main redraft; let him start slow, then buy low expectin…
It rests on these draft principles:
The books do not agree on him. James Cook is a board split: Yahoo takes him ~10, ESPN lets him slide to ~13 (3.1-pick gap). That 3.1-pick gap between Yahoo and ESPN is the practical thing to act on: in a ESPN-priced room he is available later than his consensus number suggests.
He goes at 11.9 overall on consensus, with a 3.1-pick spread between cheapest and priciest. He is latest on espn and earliest on yahoo: sleeper 12.7, espn 13, yahoo 9.9, underdog 10.8, cbs 12.1.
At 11.9 you are paying a WR2 price for a player whose median outcome is a WR2 season with five weeks that win you the game outright and four that lose it. The bust rate of 37% is real and so is the 29% boom rate.
The call: fair value, not a discount. The reason to take him is the ceiling and the betting record. The reason to pass is that you are buying a profile that has beaten its expected output four years running and only needs to stop once.
Ask the Coach about James Cook Every number in this piece, plus your league's roster and scoring, in one answer.Expected points are computed from realised usage — targets, air yards and field position — against league-average conversion, not from projections. Prop records cover 120 graded lines. Projections are anchored to consensus with our engine applied as a capped tilt; the raw engine figure is stated where it differs. Nothing here is betting advice.