Research the prop. Understand the projection.

Live DFS lines plus a sportsbook reference with the vig removed. Verified markets grade against final official statistics. New pitch and workload markets are provisional research inventory, and unverified settlement stays pending.

• Nine batter lineup through a base out state machine vs starter → bullpen
• Empirical Bayes matchup rates × platoon splits × park and weather × umpire zones
• Statcast contact, whiff and arsenal tilts, posted lineups, steals simulated during the game

Live lines from four DFS books, priced on player rates measured from recent box scores, graded nightly. Injury statuses default to healthy until a licensed feed is connected, and affected projections are flagged.

• Usage → shot/FT/turnover → rebound battle → assist/steal/block attribution
• Minutes and availability rolls per simulation, including measured coach decision rates by rotation rank. Probabilities are DNP conditional (DFS voids DNP legs)
• Combos and fantasy correlate by construction (same simulated possessions)

The engine is built and has passed two audits. Live player rates and DFS lines connect when the season opens in late October.

• Rest availability uses the prior season's excess single-game absence rate on back-to-back games, by minutes tier
• On ball usage redistribution when stars sit (∝ minutes × on ball share, conservation proven)
• Garbage time simulated: blowout sims split into competitive and bench heavy floors

Real DFS lines priced against measured NFL data: player usage and efficiency from nflverse (community play by play and weekly stats, CC BY 4.0, used with attribution), team profiles anchored to the measured league, and kickoff weather from the National Weather Service. Injury designations enter from the roster file and, when the licensed feed is configured, from RotoWire. The board states which. No sportsbook posts two sided prices on these markets, so edges read model versus line, and every grade is capped at B+ until the backtest earns more. Graded weekly from official box scores.

• Environment (drives, plays, points, weather) → pass rate adjusted per drive by the script
• Allocation by role: targets = routes × TPRR, carries = rush share, TDs = end zone equity
• Conservation by construction: QB passing yards ≡ Σ receiver yards per sim (structural stack correlations)

Real DFS lines priced against measured college data: player usage and efficiency from last season's play, opponent adjusted team ratings, returning production, and transfer moves. Rosters are current, player availability has no live feed yet, and every player builds healthy, stated on the board. No sportsbook posts these markets, so edges read model versus line, and every grade is capped at B+ until the backtest earns more. Graded volume is still thin (1,371 settled rows) and the probabilities run overconfident at the extremes, so treat them as directional until the calibration curve tightens.

• Starter pulls and garbage time are simulated (pull margins per staff, backups absorb usage)
• Tempo identities per team (55 to 85 plays per game), mismatch scale environments
• Roster volatility engine: transfer, freshman and committee inputs → grading that demotes readily

The engine is built and audited, and simulates college structure directly: five fouls, the bonus free throw ramp, end game fouling, two foul benching, and rosters heavy with transfers. Live stats and lines connect with the November season, and every grade is capped at B+ until the backtest earns more.

• Minutes come first: availability, foul trouble (disqualification on five fouls), benching on two fouls in the first half by coach tendency, blowout compression, and tournament rotation tightening are realized per simulation, so every probability is DNP conditional
• The bonus and one and one FT structure is simulated within each half (foul accumulation ramps team FT trips, and front end misses are live rebounds), and the closed form FTA claim is audited against the sim
• End game fouling is a first class channel: finishes that are close but not tied extend the game with intentional foul cycles (leading team shoots two, trailing team forces threes), so FTA and 3PA distributions carry the real late game fat tail

Live DFS lines priced on serve and return rates measured from each player's recent matches, split by surface. No sportsbook posts tennis player props, so edges read model versus line, and every grade is capped at B+ until the backtest earns more.

• Deuce games, tiebreaks (one then two serve rotation, tb10 deciders), sets, Bo3 and Bo5
• Barnett and Clarke serve and return blend over tour × surface anchors
• Player rates measured from recent matches (split by surface, weighted by recency), EB shrunk toward the anchors

The engine is built, verified by harness, and adversarially audited, with PGA cut events and LIV 54 hole shotgun events as separate formats. DFS golf lines are confirmed available. The missing piece is a licensed strokes gained data feed, not the calendar.

• Course environment comes first: difficulty per hole sets the outcome environment, realized exactly by the sim
• PGA and LIV as different structures: cut truncation at 36 holes vs a 54 hole shotgun with no cut
• The player's own simulated rounds decide his cut, so cut risk correlates with every prop

Real DFS lines priced against measured fighter statistics. Full cards build on fight days, and every published prop grades against final fight statistics the morning after. No sportsbook posts these markets in usable form across books, so edges read model versus line, and every grade is capped at B+ until the backtest earns more.

• Fight structure comes first: a finish hazard per tick yields closed form fight time, round reach, and decision claims the sim realizes exactly
• Three round and five round fights as different structures, and rounds start standing (ground spells end at the bell)
• Early finishes settle volume props, so finish risk is priced into the distribution, never voided away

Real DFS lines priced against measured team and player inputs: posted lineups with a status of confirmed or projected, match statistics per player, and substitution timing from recent matches. An unposted lineup caps every grade for that match. No sportsbook posts soccer player props in usable form across books, so edges read model versus line, and every grade is capped at B+ until the backtest earns more.

• Minutes come first: starts, timing of substitutions off, and windows for substitutions on are realized per simulation, so every probability is conditional on the player appearing (DFS voids legs for a player who never enters)
• Possession is zero sum and splits the prop world: one environment drives your passing volume AND the opponent's defensive actions and keeper saves in the same sims
• Share slots conserve team volume exactly: substitutions pass the opportunity slot while conversion rates stay the player's own

Real DFS lines priced against measured pro match data: league pace, team styles, rosters, and role shares, using only games played before each series. Projections are made before champion select, so distributions widen to carry that uncertainty. No sportsbook posts these markets, so edges read model versus line, and every grade is capped at B+ until the backtest earns more.

• Game duration comes first: an end hazard per minute drives every claim, from expected map length to P(reach 30/35) to team kills, and the sim realizes each exactly
• Early ends settle volume props (a 22 minute stomp cashes unders). Remakes, forfeits and cancellations are the void events, never simulated inside a map
• Line scope is part of the market: Map 1, Maps 1 and 2, Maps 1 to 3 and series lines sum over the maps actually played, and P(map 3 happens) is a structural channel like the golf cut

Real DFS lines priced against measured match data: kill and headshot shares per player, team strength, and map pools from recent finished maps. Before the map veto, projections mix the likely map pool and say so, and a stand in or unconfirmed roster caps the grade. No sportsbook posts these markets, so edges read model versus line, and every grade is capped at B+ until the backtest earns more.

• Rounds come first: a CS2 map is a race to 13, so every volume stat scales with rounds played, and the rounds distribution is exact score race mathematics the simulation realizes directly
• Early ends settle volume props (a 13 to 3 stomp cashes unders). Forfeits and cancellations are the void events, never simulated inside a map
• The inversion: close maps run long and lift both teams' kills, so opposing players' overs are positively related through rounds (surfaced on every relevant prop, the honest number for stacking a slip)