The nerd optimizer. QUICK DRAW is the simple one: a few dials and a spin. This is the other one. Every rule, every percent, up to 150 lineups at a time. Primetime games have their own SHOWDOWN mode up top: captain mode, one game, every legal lineup counted.
TwoGunGPT simulates every game 10,000 times. This page gets 5,000 of them: half to build, half to grade. Each player gets a MEDIAN (his normal game), a CEILING (a top-20% game, 1 in 5) and a NUKE (a top-5% game, 1 in 20). OWN is our own ownership projection, for the Milly and for small fields. Don't like a number? Change it. The projection by a percent or outright, and the ownership too.
It works from exact lists. A stack is a QB, 0 to 3 of his catchers and 0, 1 or 2 bring-backs from the other side. Nobody else from that game. No running back with his own QB. No defense facing him. For every stack, the engine finds its 4 best fill-ins under your locks, X's, games and salary, on your numbers. Exact per list: nothing in it is a guess. The other caps (the team cap, ownership, MIN and MAX, the stack caps, DIFFERS BY) are checked inside the list, lineup by lineup. You set the stack mix in percents, or let AI set it from the sims.
Those lists are the magazine: tens of thousands of lineups on a main slate. Each lineup you get comes from your slice of it (THE TWIN FLOOR, below): the best legal one there on your numbers, drawn inside the band. With RANDOMNESS on, it's the best on that lineup's own shaken numbers. That's the best lineup in an exact list. It is not always the exact best lineup: one we didn't list can beat it by a few points. Your caps work on the list, lineup by lineup. When the rules leave nothing legal, the engine digs: a fresh exact list with the players your lineups share most left out. A short list gets more fill-ins per stack. A player owed his MIN gets a list built with him locked. The result shows how big the magazine was and how many lineups each rule cut. If a build still stops, it says which rule ran out.
When a QB has a top-20% game in our sims, we count how many of his top 4 catchers have one too, averaged over every QB with a 15+ CEILING. That's the mix. Next to it the page prints what luck alone would give if nobody moved together. Same for the bring-back: for example 55% next to about 49% by luck means the other side barely moves with him. The naked QB (nobody with him) only goes to the dual-threat QBs: 25+ rushing yards or 4+ carries a game in our sims. A QB who scores with his legs doesn't need his receivers. No dual-threat QB on the slate? That share goes to one catcher. The bring-back shares are how often someone on the other side booms with him, and how often two of them do: two backs are a shootout bet, and the lineup pays up and down when it lands. At 10 lineups or more, the build hits the mix exactly, unless a shape can't fit your rules. Under 10, each lineup draws its own shape from it.
A rule says how a game goes for part of your build: "30% of my lineups: this game goes OVER", or "20%: the home team by 10+". Those lineups stack that game. Each one is built only on our simulated Sundays where that happened. That game's players are read again inside those Sundays, so a team that wins by 10+ gets its winning-day numbers. Your MY PROJ and ADJ still count: the world moves a player's number by its lift, scaled by your number the way RANDOMNESS is. Zero a player and no world lifts him. The other games don't move: in our sims, games don't move together.
We read each game's score off the sims: the offense, the kicker and the other side's D. OVER means past the game's own middle in our sims, not the book's total. Our middle often sits over the book's total, so a book cut would make OVER the usual case. Each rule shows how often our sims go its way and the number OVER cut at. Under 300 of our build Sundays, the page warns you. Under 100, it won't build. OVER, who-wins and margin worlds only. No UNDER worlds yet: the engine keeps a team's RB and D out of its own stack game, so an UNDER world can't build the lineups it means. AI fills a rule's % with the sims' own odds of that world.
The shares work like the stack mix. At 10 lineups or more the build hits them exactly. Under 10, each lineup draws its world. A world lineup your other rules can't build (the caps, a MIN, your locks, DIFFERS BY) gives way to one without it, and ASKED VS GOT says so. After the build, the STACKS tab grades each world's lineups on the held-out Sundays inside their world, on all of them and when the game goes another way, next to your lineups with no world. It's a bet. A world's lineups pay when it happens and can cost you when it doesn't. Strangers with the same rules still get their own slice of the magazine (THE TWIN FLOOR).
Points minus strength × ownership, in points. At 8, a 30%-owned guy loses 2.4 points. At 15, he loses 4.5. The same points come off any number, so it bites harder on MEDIAN than on NUKE. A player with no ownership on file gets the middle of his position near his salary, not zero. Under the lineups, THE PRICE says what your average lineup gives up to the best single lineup under your hard rules.
RANDOMNESS shakes the numbers before every lineup. At 0 it does nothing, and MOVE WITH THE SIMS greys out. With it on (the default), each lineup gets its own simulated Sunday: if a QB goes off in it, his receivers go off with him, and that lineup leans into the stack. A player's swing scales with your number: zero him and he never booms; double him and he swings double. Turn it off and every player moves up or down on his own, by up to that percent. The magazine is built on your numbers before the shake, so a shaken lineup is the best one in the list for its Sunday.
Two strangers with the same settings shouldn't get the same lineups. So each build opens only its own slice of the magazine: about 1 lineup in 25 (1 in 33 past 50 lineups), picked at random by your seed and your account. A stranger gets a different slice, so a lineup is open to about 1 build in 25 whatever the rules, until you lock 6 (below). Then each lineup is a seeded draw from the legal ones within 1% of the best in your slice (up to 3% with 3 or more locks). The slice picks lineups, not players, so it leans on nobody you didn't touch. If your slice runs short, the engine digs deeper for more. If even that leaves nothing, the build stops and says so: loosen a lock or a cap. It runs even at randomness 0. Lock 6 or more of your own and the slice gives way when your locks leave too few lineups for it: those lineups are mostly your picks, and strangers who lock the same 6 can get the same ones. The page says so when it happens. A recipe's locks don't count toward the 6. We tested it on week 5: 200 users a test, the same settings and the same locks. The two most popular players locked, 20 lineups each: the most common lineup showed up in 7.5% of the builds. 50 lineups: 7.5%. 150 lineups: 7.5%. The 4 most-owned locked, 20 lineups: 9%. Only the 2 highest-total games left in, 5 lineups: 7%. Everybody typing the same seed: 8%. It costs about 4 points of median a lineup at 20 or 50 next to a build with no floor, more at 1 lineup and at 150. Each build prints its own price: we build the same batch with the floor off, same seed, same rules, grade both on the same held-out Sundays and print the gap, with its ±. "This batch gave up 4.0 points of CEILING a lineup to stay nobody's twin." Only the number. Never that batch. Your seed replays your builds only.
The page follows DraftKings' newest main slate. Until our numbers for it land, you can set your rules but not build. DraftKings stops taking new entries at the first kickoff. Once every game has kicked off, BUILD turns off, and the page points you to the next slate and to the primetime games (this page's SHOWDOWN mode, or QUICK DRAW SHOWDOWN). A game our board hasn't priced yet stays off the board.
When two of your rules can't both hold, the earlier one wins: OUT, then locks, then MAX, then the caps, then MIN, then the mixes and SCRIPT WORLDS. A MIN on a player who's out, or whose game is out, drops. So does a SCRIPT WORLDS rule on a game that's out. A world past the stack caps takes a naked QB from its game when one may go naked; else it gives way. A lock beats every cap: a locked QB stacks past MAX LINEUPS PER STACK TEAM. Type a MIN above his MAX and the MAX moves up to it. The MAX you typed comes back when the MIN comes down. The caps beat a MIN: a QB at his stack cap isn't forced in, and a player no lineup can hold under your caps sits that lineup out. A naked QB stacks nobody, so he doesn't count toward the stack caps. Rules that can't all hold at all keep BUILD off and say why in plain words: a QB and his own RB, two D's in every lineup, more locks at a position than DraftKings has slots, locks already over a cap or the salary, QB MINs over 100%, one game left in. After the build, EXPOSURE shows ASKED VS GOT in red for every MIN, MAX or stack cap that missed, and the rule that beat it.
Two numbers under the lineups, after they land. The give-up: how far your average lineup sits below the best single lineup under your hard rules (your locks, X's, games, salary, the team cap and the ownership caps), on your numbers, over every stack shape. MIN, MAX and the stack caps don't apply to one lineup. It doesn't say why: we don't split that gap by cause. Only the twin floor gets its own price, measured on its own. The twin floor's price: see THE TWIN FLOOR.
Up to 150. 150 lineups grade about 5 points of median a lineup lower than 20 on week 5 (141.0 vs 145.8): the more you build, the deeper into the magazine they go. The magazine loads first, then the lineups come one at a time. A player at his MAX exposure drops off the list for the rest of the build. A player under his MIN is spread across the build by a seeded coin, then forced in when the build runs short of lineups to catch him up. Percents round to the closest lineup: 30% of 8 lineups is 2. Every pair of lineups differs by at least the number of players you set.
DOWNLOAD FOR DK makes DraftKings' upload file. In each lineup, FLEX holds the latest game of its position, so a late swap has room. The file name carries the week, our board and a batch tag. Never your seed.
Our MEDIAN, CEILING and NUKE come from all the sims. On this page the sims split in two. The first half shakes the lineups and sets the AI mix. The second half only grades: 2,500 Sundays that never shook a lineup or set a mix. Every grade, the X-RAY and the PORTFOLIO read those. It's our model checking itself on fresh draws, not on real games. Grades use our sims, not your numbers. Your MY PROJ and ADJ move the build, not the grade.
Type a score. Each lineup's CLEARS column is the share of the held-out Sundays it hits that score or more. The default is 180. Sort by it, or by MED, CEIL or NUKE. The DK file always keeps build order.
2,500 Sundays is a lot, not infinite. So every grade carries a ±: two standard errors, about a 95% range. Lineups the sims can't tell apart share a BAND. We test each gap on the same Sundays (paired), and a gap inside its ± is noise. Inside a band the order means nothing, so they stay in build order. WOULD THE ORDER HOLD? splits the Sundays in half, ranks your lineups on each half and checks they agree (Spearman, stepped up to all 2,500 with Spearman-Brown). STEADY ORDER is 0.9 or better. Under 0.7 it's NOISE. Open a lineup and set it against any other: the page says "the gap is real" or "can't tell with 2,500 Sundays", and about how many Sundays it would take.
Tap a lineup. Its 2,500 held-out Sundays as a histogram, p10 to p99 each with a ±, and the chance it scores any number you type. Then what its best 5% need. The QB: how it scores when he has a top-20% game and when he doesn't. The game script: each game's score is read off the sims, and we look for the margin and total its best Sundays lean on. OVER and UNDER cut at the game's own middle in our sims, not the book's number. A script shows only when it clears its noise. Each player's LIFT: how much more he scores on its best 5% than on an average Sunday, and his SHARE of the lineup's lift. THE HINGE: pin a player at his middle game and see how much the best 5% lose. Biggest loss = the hinge.
Lineups that need the same game rise and fall together. So 50 of them can play like a handful. HOW MANY THEY PLAY LIKE is the number of lineups that never move together that would give the same chance that at least one clears the bar. We print it next to that chance. More isn't always better: spread-out lineups can each be weaker. The coverage curve adds your lineups in build order and sets them against lineups that never move together. The heat map shows which pairs move together, sorted by stack game. No contest field here yet: the bar is a score you pick.
Sims, not results. THE FIELD tab drops your lineups into a fake Milly. It has as many lineups as the newest real Milly on file had (the tab names the week and the size). They're drawn from our Milly ownership. They stack the way this season's real Millys stacked: QB + 1 catcher and no bring-back in about 29% of them, no stack at all in about 11%. They spend the way the real ones spent: half use the whole cap, and about 2% leave 800 or more on the table. Each player's share of the fake field lands, on average, within about half a point of our ownership (a few miss by 2). We score the whole field on the same held-out Sundays your lineups are graded on. Then we keep the score each line took: cash, top 1%, top 0.1%, first, and every few percent of the field between. Your lineup's score that Sunday, set against those lines, says where it finished. UNDER 200 plays the same Sundays in a 150-entrant GPP drawn from that fake Milly: TOP 3 = your lineup lands in the top 3 of 150 (a finish tier, like the Milly's TOP 1%), CASH = the top 35, FINISH = its expected place; a real field under 200 owns more chalk than a Milly, so switch FIELD to SMALL FIELD when it's posted and read the Milly-field numbers as the easier case. CASH and TOP 1% count a tie at the line, because DraftKings splits a tied place's prize. FINISH is the share of the field at or above you: top 21% beats 79% of it. CASH, TOP 1% and FINISH each have a ±, two standard errors over the Sundays. The cut lines and the good and bad Sundays are reads off the same Sundays, not estimates, so they carry none. Top 1% is rare, so its ± is wide next to it. The field is built from our ownership, not your MY OWN. It's the same for everybody. It grades your build. It never builds it, and nothing here enters a contest. It's rebuilt every time our numbers or our ownership move. Until it is, the tab says the field is rebuilding and shows no rates. What it gets wrong: real fields repeat lineups. About 10% of a real Milly's entries share their lineup with someone. In ours, about 1% or less do (the tab prints this week's). So real chalk splits more prizes than it does here. And our sims aren't the real Sunday. No money, no ROI, no win odds here.
Signed in, your settings and your last build ride on your account. Open the page on any device and they're there.
The link carries your settings. Not your seed, so whoever opens it gets their own batch. It carries 5 locks at most, and 5 of your own numbers. They still make two changes of their own before they build.
Their terms say a lineup tool has to make you make real decisions. Lock, exclude, cap or re-project at least two players before you build. Only players on the board count, and only changes big enough to matter: ADJ 5%+, MY PROJ 1+ point off ours, MY OWN 3+ points off ours (or a 20% move worth a point), a MIN or MAX worth a lineup. A SCRIPT WORLDS rule counts as one when it's worth a lineup. A rule on AI doesn't: those odds are ours. The same world twice counts once. Changes and rules that came in someone's recipe don't count. Then it's yours.
Sims, not results. No lineup here is a prediction.
THE GUNSMITH for DraftKings showdowns: one game, a captain at 1.5x points (and 1.5x salary) plus five flex. Pick any primetime game that hasn't kicked off. Want it simple? QUICK DRAW SHOWDOWN calls the game and spins.
Only in showdown can you count every lineup. We list every legal one: a captain and five, both teams, $50,000 or less, $45,000 or more. That's the space. Every rule you set filters it, live: "your rules leave 1,912". The PLAYERS tab shows each player's share of what's left, as captain and in the flex.
Our sims play the game 5,000 times. Half build, half grade. The build reads each lineup's average and spread on the build half, with the captain at 1.5x and the way the six move together. MEDIAN is the average. CEILING adds 0.84 spreads (about a 1-in-5 night). NUKE adds 1.64 (about 1 in 20). We checked it against the exact numbers on 8 games: the same top lineup 18 times in 24, and the top 50 by our numbers score within half a point of the exact top 50. The numbers themselves sit a point or two off (NUKE about three under). The grades are exact, on the other half. The sims never grade their own homework.
SPLIT × CAPTAIN SIDE = how many from each team, and which team the captain plays for. THE STARTING QBs = both, one, or neither. AI = the sims' own mix: the shape of the best possible lineup on 400 of our simulated games (house rules on). They're targets, not promises. A lineup that can't fit its target swaps with a later one first, then takes the closest shape. THE MIX tab shows asked vs got. Why not exact? Exact mixes squeeze everyone into tiny corners of the space, and strangers land on the same lineups (measured: 43% to 73% of builds).
No kicker or defense at captain by default: they've been this season's worst captains (the receipts are on SHOWDOWN ROI). Turn them on, or give one a CPT MIN.
Each % of our showdown ownership costs the lineup strength/100 points: captain % on the captain, flex % on the five. At 8, a 30%-owned flex loses 2.4 points.
OUT > lock > MAX > salary and captains > MIN > the mixes. A MAX is never passed. When a lineup can't be filled, the mixes give way first, then the smallest MIN, then DIFFERS BY by one spot (with a note). Then the build stops and names the rule that ran out.
Strangers with the same settings must not get the same lineups. Each build opens its own slice of the space, about 1 lineup in 33, drawn from your seed and your account. Then each lineup is the best one open, give or take 1%. Our test: 200 strangers, the same settings, 17 setups at 1, 5, 20 and 50 lineups. The most common lineup was in 8% of builds at worst. It costs points, and every build says how many. Lock too much and the slice can run dry: the build stops and says so.
Every lineup in the space is a dot. Up = more points on the build Sundays (your BUILD ON number). Right = more owned (or more salary, before our ownership is posted). Gold = your rules leave it. Green = your build. The dots use the build's numbers: close, not exact. Your lineups' grades are exact.
CPT, then the five FLEX. The CPT column carries DraftKings' own captain ID, so the upload page takes it as is. We never enter anything for you.
Their terms say a lineup tool has to make you make real decisions. Make at least two before you build: a lock, an X, a CPT or FLEX MIN or MAX, your own mix, the captain rules, a salary range. Each counts only when it moves at least one lineup against the lineups a build would reach without it. The AI's mixes and the dials don't count.
Sims, not results. No lineup here is a prediction.
TwoGunGPT's free advanced DraftKings NFL lineup optimizer. You set the rules, down to every percent, and it builds up to 150 lineups on TwoGunGPT's own projections, ownership and simulations.
QUICK DRAW is a few dials and one lineup a spin. THE GUNSMITH builds a whole batch: exposure caps, stack mixes, bring-backs, ownership, randomness and script worlds. If you enter a lot of lineups, use THE GUNSMITH.
The DraftKings NFL Sunday main slate (classic) and the primetime showdowns (captain mode), once DraftKings posts their salaries.
DraftKings' rules for lineup tools ask you to make real decisions first, so people aren't handed the same lineups. Make at least two real changes, like a lock, an X or a cap on a player. The dials and the AI's mixes don't count.
On 2,500 simulated Sundays the build never saw. Every grade carries a ± so you can tell a real gap from noise. Sims, not results.
Yes. It needs a free TwoGunGPT account, the same one QUICK DRAW and the Ceiling Board use. No credit card. Your settings and your last build ride on the account.