Scores View all
FRI 6:00 PM ET
ISU
UCF
—
—
FRI 6:30 PM ET
BROC
MS
—
—
FRI 7:30 PM ET
CLEM
DUKE
—
—
FRI 7:30 PM ET
BGSU
TOL
—
—
FRI 8:00 PM ET
UTEP
AF
—
—
FRI 8:00 PM ET
ORE
MSU
—
—
FRI 9:00 PM ET
UNM
WYO
—
—
SAT 12:00 PM ET
LIU
RM
—
—
Painterly illustration of a quarterback, running back and receivers under stadium lights for FSN’s Week 9 DFS report card.
NFL 2026 NFL Season

FSN DFS Week 9 Report Card: Small Bias, Big Misses

FSN reviews its older archived Week 9 DFS forecasts: 295 DraftKings and 294 FanDuel matched records, position-level errors, notable hits and misses, and range coverage.

FSN newsroom article

Produced through an FSN-managed editorial workflow using sourced reporting, research and publication tools; subject to FSN editorial and corrections standards. How FSN uses AI & automation Corrections

FSN Football Sports News1717 words · 8 min read
Share X f ✉
In this story8 sections
  1. Week 9 DFS projection accuracy: the full matched sample
  2. Quarterbacks: role assumptions mattered more than a small average correction
  3. Running backs: Walker surged; Swift and Jacobs fell short
  4. Wide receivers: Boutte’s breakout was not inside the forecast
  5. Tight ends: the lowest absolute error is not a blanket passing grade
  6. The saved ranges need scrutiny, not a victory lap
  7. Methodology, exclusions and what this report does not claim
  8. FSN takeaway

Kayshon Boutte’s saved FSN projection was 4.47 DraftKings points. His Week 9 result was 31.50. D’Andre Swift went in the opposite direction: a 25.31-point forecast and a 3.10-point result. Those two misses explain why a model can look reasonably balanced in the aggregate while being badly wrong about individual players.

Across the matched Week 9 sample, FSN’s older archived forecasts overpredicted scoring by an average of only 0.56 DraftKings points and 0.69 FanDuel points. But the average absolute miss was 4.94 and 4.52 points, respectively. Positive and negative errors canceled each other in the first measure. They did not cancel each other in a lineup.

Scope of this retrospective: This report evaluates preserved pre-kickoff forecasts from FSN’s older archive, not the current projection model. Exact equivalence between those older records and every number displayed by the public builder was not recorded. Results below cover 295 matched DraftKings records and 294 matched FanDuel records, not every player in the archive or a specific contest’s starting lineup.

Week 9 DFS projection accuracy: the full matched sample

2026 Week 9: older archived forecasts versus source-calculated fantasy scores
Platform Records scored Mean absolute error Bias: actual minus forecast Root mean square error
DraftKings 295 4.936 -0.556 6.970
FanDuel 294 4.519 -0.685 6.471

All error figures are fantasy points. Mean absolute error measures the average size of a miss without allowing an underestimate to offset an overestimate. Bias measures direction: negative values mean the forecasts ran high. Root mean square error gives larger misses more influence.

The gap between absolute error and signed bias is the central finding. This was not simply a model that needed every forecast shifted down half a point. Large errors existed in both directions. A uniform adjustment would not have repaired a near-zero forecast for a quarterback who played a full game, or anticipated an explosive receiving performance.

These are equally weighted player-record comparisons, including low-usage players and recorded zero-point outcomes. They are not salary-weighted, limited to popular starters, or a measurement of contest winnings. FanDuel’s lower absolute error also does not establish a superior model: the scoring system and included sample differ.

Quarterbacks: role assumptions mattered more than a small average correction

Shedeur Sanders is the clearest review priority. The archive assigned him 0.05 points on both platforms; his source-calculated result was 22.68. The official Browns-Saints box score records 297 passing yards, two passing touchdowns, no interceptions and 28 rushing yards on 31 pass attempts and 10 carries.

That is not a small efficiency miss around a plausible workload. The archived number was effectively forecasting no fantasy contribution. It makes quarterback participation and starter assumptions the first things to investigate. It does not, by itself, prove whether the cause was an injury update, a depth-chart input or another part of the old model.

Jordan Love exposed a different weakness. His saved forecasts were 18.18 DraftKings points and 19.44 FanDuel points; his calculated result was 39.18 on each. The Packers-Patriots box score shows 372 passing yards and three passing touchdowns, plus a 13-yard rushing touchdown. The scoring summary also records his successful two-point run.

Sanders and Love should not be filed under one generic explanation. Sanders’s forecast demands a role review. Love’s result tests the model’s ability to represent a high-scoring outcome. The quarterback sample had the largest absolute error on both platforms, although its 31 records include more than just starting quarterbacks.

Running backs: Walker surged; Swift and Jacobs fell short

Kenneth Walker III generated 39.30 DraftKings points and 37.30 FanDuel points against saved forecasts of 18.60 and 17.86. The official Jets-Chiefs box score records 102 rushing yards and a rushing touchdown, plus four catches for 41 yards and two receiving touchdowns. His scoring reached well beyond either archived ceiling.

That is an upside miss, not automatically a successful prediction because a productive player appeared in the pool. A fair audit credits a close forecast and records a large underestimate as a large underestimate, regardless of whether the player helped some lineups.

Swift supplied the largest overprediction in the matched sample on both platforms. His 3.10-point result fell 22.21 points below his DraftKings forecast and 21.72 below his FanDuel forecast. Chicago’s official box score shows 12 carries for 31 yards, with no receiving production or touchdown to offset the modest rushing total.

Josh Jacobs was another substantial miss: 20.76 projected versus 0.50 actual on DraftKings, and 17.75 versus 0.50 on FanDuel. His four carries for five yards are documented in the Packers-Patriots box score. The result identifies a workload-sensitive miss; the archive alone does not establish why that workload occurred.

There were close calls in the good sense, too. Breece Hall’s 61 rushing yards and three catches for 26 yards produced 11.70 DraftKings points and 10.20 FanDuel points. His forecasts were 11.52 and 10.02: an underestimate of just 0.18 on either platform. One accurate example does not erase the position’s broader error, but it belongs in the record alongside the misses.

Wide receivers: Boutte’s breakout was not inside the forecast

Boutte delivered the largest positive error in the matched sample: 27.03 points above his DraftKings projection and 26.01 above his FanDuel projection. Houston’s official box score records four receptions for 125 yards and two touchdowns.

His archived DraftKings range ran from 2.68 to 8.08 points; the FanDuel range ran from 2.20 to 5.93. The result exceeded both. Calling this a “ceiling hit” would confuse a player having a high-scoring game with the model successfully anticipating that outcome.

Amon-Ra St. Brown provides the counterexample. His eight catches for 86 yards, confirmed by the Lions-Dolphins box score, produced 16.60 DraftKings points against a 16.58 forecast. The FanDuel comparison was less exact but still close: 12.60 actual against 14.04 projected.

The receiver sample’s positive signed bias was small: 0.337 DraftKings points and 0.183 FanDuel points. That does not make the individual forecasts uniformly reliable. St. Brown’s near-match and Boutte’s major underestimate coexist inside those same averages.

Tight ends: the lowest absolute error is not a blanket passing grade

Tight ends had the lowest raw-point mean absolute error of the four positions: 4.027 on DraftKings and 3.496 on FanDuel. That comparison needs context. A position with smaller typical scores can have smaller point errors without being more accurately forecast in proportional terms.

Dallas Goedert was a clean near-match. His five catches for 50 yards in the Giants-Eagles box score produced 10.00 DraftKings points against 10.18 projected, and 7.50 FanDuel points against 7.76 projected.

Tucker Kraft went the other way. One catch for one yard and eight rushing yards produced 1.90 DraftKings points and 1.40 FanDuel points. The saved forecasts were 12.84 and 10.38. That is a meaningful miss despite tight ends having the smallest position-level absolute error.

Position-level mean absolute error, in fantasy points
Position DraftKings records DraftKings error FanDuel records FanDuel error
Quarterback 31 6.995 31 7.112
Running back 77 4.983 76 4.597
Wide receiver 119 4.889 119 4.377
Tight end 68 4.027 68 3.496

The saved ranges need scrutiny, not a victory lap

Of 291 DraftKings records with saved ranges, 114 results landed inside them: 39.18%. Another 122 finished below the saved floor and 55 above the ceiling. FanDuel had 112 inside out of 290, or 38.62%, with 125 below and 53 above.

The model labels those boundaries as the 25th and 90th percentiles. A calibrated interval with those boundaries would nominally cover about 65% of outcomes, subject to discrete scores and boundary conventions. These observed rates are a reason to investigate, not evidence that the ranges were well calibrated.

One selected week cannot establish long-run calibration. Players share game environments, the matched sample is incomplete, and these records belong to the older archive. The appropriate next test is consistent coverage across additional weeks and player groups, not changing the old ranges after seeing the results.

Methodology, exclusions and what this report does not claim

We inspected all five archive pages for each platform. Each platform’s archive contained 467 saved records with stored capture times preceding kickoff. Thirty defensive entries per platform fall outside this offensive audit, leaving 437 offensive forecasts per platform.

DraftKings matched 295 of those offensive forecasts, or 67.51%. Its exclusions were 141 unavailable stat matches and one unresolved player identity. FanDuel matched 294, or 67.28%, excluding 142 unavailable matches and one unresolved identity. Missing stat rows were not assigned zero. Explicit zero scores in matched source rows remained in the calculation.

Actuals are FSN’s calculations from imported player statistics, using the DraftKings NFL Classic rules and FanDuel’s published NFL scoring, including applicable bonuses and all-phase lost fumbles. The nflverse player-statistics documentation describes the source fields. The scoring cache was refreshed on November 12, 2026, Mountain Time; official team box scores were checked for the examples discussed above.

The full matched sample relies on those imported statistics; it has not been independently reconciled row by row against settled contest totals. Source-complete scoring is not certification of platform settlement, and later stat corrections can change the comparison. All figures here use the reader’s full-score calculations, not its separately labeled provisional metrics.

Forecasts are saved means, not medians. Historical tool pages can recalculate, so today’s values on the Week 9 DFS builder were not substituted for pregame records. The archive did not document exact parity with the old public builder, and stored timestamps are not an independent external attestation. This is a retrospective, not a backdated prediction article.

FSN takeaway

The useful conclusion is not a letter grade or a cherry-picked hit rate. The older Week 9 forecasts showed relatively small directional bias, sizeable individual errors and limited observed range coverage. Role assumptions deserve attention first, while headline accuracy claims need the sample definition and exclusions attached.

For current research, use the FSN NFL DFS Lineup Builder. Earlier coverage includes the Week 2 quarterback audit and Week 2 RB/WR report card. Those selected-player reviews use different scopes; they are not directly comparable to this full matched-sample report.

NFL data via nflverse (CC BY 4.0) · College data via CollegeFootballData.com