Each player carries per-game rates rather than points. Those rates are scaled by how many points his offence is projected to score that week, touchdowns moving in step with the team total and yardage at a slower rate, on the reasoning that a team projected for thirty rather than twenty is finishing more of the drives it was already having rather than moving half as much ball again.
The result is scored under this league's settings, including the yardage bonuses. Those are paid as probabilities rather than all-or-nothing: a seventy-yard receiver who alternates twenty and a hundred and twenty clears the hundred-yard bonus about half the time, and a metronome at seventy never does. Their expected points are not the same.
Every remaining week is projected separately, because byes and opponents differ. Then twenty thousand seasons are simulated against the real schedule with this league's total-points tiebreaker, and the finishes are counted. That last part has not changed all season; what feeds it has.
Lineups are read as set. Where a slot is empty, the best eligible player on that bench is assumed into it. A player on bye is not a zero, he is simply absent, and his manager starts somebody else.
The per-player projection driving these numbers has never been tested against a completed game, because no game has been completed. Its scoring arithmetic is verified to the cent against 19,422 historical player-weeks, so the model computes what it claims to compute; whether what it claims is right is a separate and open question.
The Gazette therefore disbelieves its own model on purpose. Every projected gap between two managers is regressed a quarter of the way toward the league average before it reaches the table, on the reasoning that an unproven forecast reports a wider spread than really exists. Without that correction this issue would have opened Tommy at 93 percent rather than 86. The correction relaxes as real scores arrive and the model can be marked against them.
A player the NFL report has out, doubtful, on injured reserve, suspended, or on the physically-unable-to-perform or non-football-injury list is not projected at all — not zero points, but absent, the same as a bye. The paper's standing assumption is that a manager sets his lineup, so it seats somebody healthy in that slot rather than leaving a man who is not playing in it. Doubtful counts as absent on purpose: about one doubtful player in four suits up, and nobody starts those odds when the bench has an alternative.
Questionable is a haircut, not a zero: 85 percent of the projection, because most questionable players do play, some on a reduced load, and a manager who sees a Sunday downgrade can still swap.
An absence moves other numbers too. A club does not simply lose what a missing player would have produced — 65 percent of it moves to whoever is left, and the rest is offence that stops happening. Remaining runners and receivers are lifted in proportion to what they were already projected for, capped at 30 percent so one absence cannot turn a fourth receiver into a star. The quarterback throwing to backups loses ground more slowly, at 35 percent of the share that went missing. When it is the quarterback himself who is out, his pass-catchers come down 12 percent. Only the top of a depth chart counts toward any of this — one quarterback, three backs, four receivers, two tight ends, ranked by the model's own numbers — because otherwise somebody's fifth receiver is always on a report and every club in the league drifts by a percent for no reason.
The report is re-read every half hour during games rather than once a morning, so a Saturday downgrade reaches Sunday's projections.
Model: gridiron (nflverse + team-score model); no Sleeper data