Week 9 Main (11 games)

Sim draw: score = max(0, proj_eff × (1 + σ_pos × z)), z ~ standard normal.
Variance σ by position: QB 0.32 · RB 0.38 · WR 0.45 · TE 0.45 · DST 0.55.
Injury adjustments: Questionable ×0.7 · Out = removed from pool.
Weather: session-entered sustained wind of 15+ mph applies ×0.90 to QB/WR/TE; gusts of 25+ mph apply ×0.85 instead. Rain is flagged for review without an invented automatic penalty.
Stack correlation: each pass-catcher is scaled by 1 + 0.35 × correlation_weight × (qb_draw − qb_proj)/qb_proj, clamped to [0.4, 2.2].
Model settings: Results Range and Simulation Range scale player-outcome variance. Projection Weight adds preference for raw projection; Ownership Weight follows the contest field (leverage in large fields, a milder blend in small fields, chalk in single entry); Upside Weight rewards simulated outcomes above projection. Chase 3-year upside rewards verified 25+ point rates and penalizes sub-10 rates for covered players; missing players remain neutral. All settings feed the next run.
Optimizer (per sim): greedy fill of QB/RB×2/WR×3/TE/FLEX/DST under the $50K cap — picks the best jittered draw that leaves a feasible cap completion, then 15 random single-swap improvements.
Win rate: share of sims in which a lineup was the optimizer's best build.
Sets mode: the highest-win-rate lineup becomes the core; the remaining entries are deliberate 2-for-2 and 3-for-3 swaps branching off it. Every incoming player is eligible for the vacated slot, the swap stays under the cap, and locks, fades, injury statuses, and the active stack all still hold. Incoming players are ranked by sim exposure, variants alternate 2v2 and 3v3 and are deduplicated against the core and each other, and any shortfall is filled with the next-best independent sim lineups. Copy and the DK CSV export exactly the set shown — one portfolio built around a single core read, instead of unrelated lineups.
Scoring: DK NFL: 1 PPR, 1 pt / 10 rush-rec yds, 1 pt / 25 pass yds, bonuses at 100/300 yds
Exposure: share of sims a player appeared in the optimal lineup.
Vegas tilt: each team's implied total comes from the line — impl = total/2 ± spread/2 (FanDuel Research + Covers + multi-book board, pulled Nov 4; lines move, re-check Sunday). Player projection is multiplied by 1 + 0.5 × (team_impl − slate_avg) / slate_avg, clamped to [0.85, 1.15]; DSTs use the inverted opponent-implied version. The FPPG baseline is opponent-blind, so this is a deliberate half-weight nudge toward scoring environment, not a re-projection.
Game scripts: five scenarios per game (blowout / shootout / close / slugfest / upset) with probabilities from a normal model (margin SD 13.5, total SD 13.0). Per-scenario player multipliers are fixed heuristics (e.g. shootout: pass-catchers ×1.15, DSTs ×0.75) — useful for thinking about game flow, not predictions. Scenario odds aren't mutually exclusive.
Locks, fades, and stacks: locked players stay in every lineup; faded players are removed; a selected 2-man or 3-man team stack requires that team's QB plus one or two linked pass-catchers. Impossible combinations return a clear error.
Session runs: every sim is kept in memory while this page stays open (max 20; rapid re-runs coalesce). The History tab reloads a run's lineups, controls, and exposures, and diffs two runs' exposures.
Limits: projections are DK FPPG baselines from the Nov 4 pull — a starting point, not a full projection model — and ownership is a salary-tier estimate rather than a real projection, so treat leverage as directional. Statuses are DK's as of the pull; re-check inactives Sunday and re-run after toggling.