2026 NFL · WEEK 2
Behind the numbers.
Projected production. Independent player priority. Measured against real results.
Start with current football data
Week 2 uses current roster identities and depth charts, Week 1 player and team statistics, offensive snaps, and play-by-play opportunities near the goal line and in the red zone. Historical usage and efficiency provide a starting point when current-season evidence is limited. A player no longer needs a preseason workbook row to receive a forecast.
A fitted target-share model uses recorded offensive participation, targets, carries, depth position and competition from teammates. Its estimate is blended equally with the prior usage allocation. Players without a recorded appearance in their team’s previous game retain the prior role estimate; missing data are not a zero workload. Team attempts, carries and targets are allocated across available players. Receiving yards, receptions and touchdowns reconcile with the projected passing offense. New players without NFL history use broad historical workload and position-efficiency priors.
Questionable and Doubtful are informational tags. Only confirmed Out, Injured Reserve and suspended statuses set points to zero. The update runs once daily; forecasts for a team are locked when its game starts.
Projections and rankings answer different questions
Projections estimate average fantasy production from a stat line. Rankings express player priority. The ranking model is trained for player ordering separately from the projection model’s point-error objective.
The selected ranking score gives 0% weight to a modeled typical scoring outcome, 85% to expected points and 15% to recent recorded scoring history. Players without prior appearances use the current forecast alone. Typical outcomes account for touchdown uncertainty; they are not guaranteed floors.
Within each position, the ranking score sets the order. Overall rankings compare that score with the next player outside a reference 12-team lineup: 1 QB, 2 RB, 2 WR, 1 TE and 1 FLEX. Superflex adds one slot eligible for QB, RB, WR or TE, changing positional demand. It does not change anyone’s projected stats or points.
Historical projection check
The target-share model was fitted on 2023 and 2024 games, with later 2024 games used to choose its settings. The frozen revision was replayed on 2025 Weeks 2–18. Because those results were examined during earlier revisions, this is a regression check, not a new independent test. Improvements over the previous release are small. Each forecast uses earlier games, prior-week rosters and pregame depth information. The fixed sample takes the top 30 players at each position by their prior eight-appearance scoring average. 1,816 of 2,040 player-games matched final statistics; 224 remain ungraded.
| Scoring | Model avg. miss | Baseline avg. miss | Model RMSE | Baseline RMSE |
|---|---|---|---|---|
| Half-PPR | 5.58 | 5.94 | 7.19 | 7.52 |
| PPR | 5.99 | 6.36 | 7.68 | 8.02 |
| Non-PPR | 5.22 | 5.57 | 6.79 | 7.11 |
The baseline is the average over up to eight earlier recorded appearances. RMSE gives larger misses more weight. The production component formula was replayed; these are not scores from the old preseason workbook model.
Historical ranking check
Ranking weights were selected using 2024 player ordering, then evaluated on 2025. The comparison asks which of two players at the same position scored more in the same week. Actual ties are omitted and predicted ties receive half credit. These small gains do not establish a large or lasting advantage.
| Scoring | Ranking order | Projection order | Pairs |
|---|---|---|---|
| Half-PPR | 61.84% | 61.51% | 23,479 |
| PPR | 62.11% | 61.90% | 23,470 |
| Non-PPR | 61.17% | 60.82% | 23,444 |
What this check can establish
The historical test does not reconstruct every pregame injury report. Missing stat lines remain ungraded, and historical weekly roster/depth snapshots have timing limitations. The Week 1 results page continues to grade the original committed pregame forecasts; it has not been replaced with a revised retrospective forecast.
No expert-site projections or rankings are model inputs. A higher projection from another provider is a reason to inspect assumptions, not proof of greater accuracy. No current betting, weather or live route-participation adjustment is included. Recorded offensive snaps inform the target-share estimate. Snaps are not routes run. The carry and team-volume formulas retain their previously tested settings.
Scoring and sources
4/pass TD, 0.04/pass yard, -2/interception, 0.1/rush or receiving yard, 6/rush or receiving TD, -2/fumble lost, 2/two-point conversion, 6/return TD, plus 1/0.5/0 per reception.
Statistics, roster, depth and play-by-play: nflverse · CC BY 4.0. Data availability and timing. Status facts: ESPN injury reports. Reproducible validation record.