Projection model
Usage, efficiency, and the number on the board.
Skill-player projections are one product. Nothing in the formula is a hidden simulation.
projection = usage × efficiency × environment × modifier
NFL
Usage is expected opportunities: dropbacks for quarterbacks, carries plus targets for running backs, targets for receivers. Seventy percent comes from a recency-weighted three-week window and thirty percent from the season rate. Efficiency is fantasy points per opportunity, pulled 35% toward the position mean on the slate so a two-target week does not set the price. Environment is the implied team total divided by 22.5, clamped so a 30-point environment cannot double a projection. The modifier is red-zone touch share against the position. DraftKings scoring is full PPR with the 100- and 300-yard bonuses. FanDuel and Yahoo use half-PPR.
Defenses do not have opportunities. Yardline blends recent team defense points with the opponent's implied total run through the DraftKings points-allowed chart, plus that defense's sacks and takeaways. A higher opponent total lowers the defense.
NBA
Minutes times points per minute, times pace relative to 100, times the implied team total relative to 114. Out zeroes the projection. Doubtful keeps 30% of the minutes. Questionable keeps 82% and flags the row for a late swap, as does a jumpy minutes log. DraftKings adds the made-three bonus and uses its rebound, assist, stock, and turnover weights. FanDuel and Yahoo use the FanDuel weights.
Badges
🔥 is the top slice of projection at that position. 💰 is the top slice of value. 👎 is the bottom slice of matchup rating, where a lower DvP is a tougher draw. The defaults are 10%, 12%, and 12%. A player can wear more than one. The slate, the lineup card, and the CLI tags column use the same cutoffs.
50/50s, Double-Ups, and tournaments
50/50s & Double-Ups: about half the field wins, so we maximize the most likely score. Tournaments: only the top ~20% pay, so we chase ceiling. Projection is lower, upside is bigger. 50/50s & Double-Ups are the default. The solver maximizes 70% of the floor and 30% of the projection, ignores ownership, and leaves questionable and doubtful players out of the pool. It starts with the stack off and three lineups. Tournaments (GPPs) maximize ceiling plus a leverage bonus, minus a small ownership penalty. A lineup's ownership percents must add up to 165 or less, and at least one player must be under 10%. They start with a quarterback stack, a bring-back, and twenty lineups. The same switch is --contest cash or --contest gpp on the command line.
Ownership is an estimate
There is no contest-reported ownership. Inside a position, the score is projection rank (36%), salary value (28%), implied team total (22%), and recent trend (14%). Up beats flat, and flat beats down. A softmax turns that into shares, scaled so the position sums to the roster spots it usually fills. Leverage is ownership rank minus ceiling rank. Positive means the ceiling is ahead of the estimate. Treat both as a hypothesis, not a field report.
Where the numbers come from
NFL stats are nflverse: weekly player stats, snap counts, injuries, the schedule, and red-zone touches counted from play-by-play. Salaries come from the contest CSV, or from the DraftKings slate when that file is loaded. Vegas totals feed the projection, the ownership estimate, and the matchup rating behind the fade badge. Until a fresh book line arrives, the totals already on the slate stay in place. NBA Out tags can refresh from the official injury report in the 90 minutes before tip. A new Out is a late scratch and leaves the pool.