actfap

Craps Odds & House Edge

The house edge on every craps bet, derived from the dice, and how much taking odds behind the line cuts it.

This prices the bets themselves from the dice, on two measures (per bet and per roll of the dice), on both sides of the line with odds behind them, and including the two different commission conventions a buy or lay bet can be charged under, plus a blended per-hour figure for a setup of several simultaneous bets. It does not predict what any individual session will do.

Your setup: what it costs you an hour

Every figure below prices one bet at a time. A real player at the table is rarely making just one: a line bet with odds, plus a place or two working behind it. This combines whichever bets you’re actually making into one blended figure, and turns it into a cost per hour at a stated pace of play, rather than leaving you to add several percentages that cannot be added directly.

135 rolls an hour at 7 players. Source: Jim Kilby, Casino Operations Management, published in Wizard of Odds’ Ask the Wizard #136 (retrieved 2026-09-16). More players slows the rate, it does not change what a single roll costs.

Place, buy or lay each point pair

Each row prices one number of the pair, since 4 and 10 (and 5 and 9, and 6 and 8) cost the same to bet, exactly like the table above. Working both numbers of a pair at once costs twice what that row shows.

Not working

Not working

Include the line bet above

Blended cost of your setup

$24.41an hour · 0.89% of your action

In action an hour$2,748.27Blended edge0.89%Cost an hour$24.41

$2,748.27 in action an hour, across 2 bets.

Where it goes, costliest bet first

Place 6 or 8 1.52% edge$18.75/hr

Pass line 0.37% edge$5.66/hr

This is the cost of the setup you asked about, not advice on how to make it cheaper. Every leg here loses money on average, and the blended figure is here to show you the total, not to help you optimize around it. The pace of play is a published average for a table your size, not a promise about your table: a faster or slower game moves the dollar figure without moving the percentage.

Pass line + 3-4-5x, blended edge

0.37%

Flat

1.41%

Per roll

0.11%

Bar is the blended edge as a share of the same bet with no odds behind it (1.41%). Average total wager per unit on the line: 3.78.

What that costs to hold, on a $10 pass line

  • Point 4 or 10

    $40 at risk

    Back with $30 to win $60

  • Point 5 or 9

    $50 at risk

    Back with $40 to win $60

  • Point 6 or 8

    $60 at risk

    Back with $50 to win $60

At the same 3-4-5x, backing the pass line blends to 0.37% with $40 to $60 on the table once a point is set. The other side blends to 0.27% with $70. The don't pass prints the lower percentage, and it is the side with more money on the table. The percentage is a share of everything you have bet, not of your line bet, so it falls when the denominator grows.

Pass line

0.37%

$40 to $60 on the table

Don't pass

0.27%

$70 on the table

Buy and lay bets: the commission is the whole cost

A buy or lay bet pays true odds, so the fair part of it cancels exactly and every cent the house takes is the 5% commission. That makes two things decide what it costs you, and no published table prints either: when the house collects the vig, and how the round-down falls against your bet size.

$20 bet, $1 vig charged only on a win

Buying: you need the number before a seven

Buy 4 or 10 $20 at risk1.67%

Buy 5 or 9 $20 at risk2.00%

Buy 6 or 8 $20 at risk2.27%

Laying: you need a seven before the number

Lay 4 or 10 $40 at risk1.67%

Lay 5 or 9 $30 at risk2.00%

Lay 6 or 8 $24 at risk2.27%

Collected only on a win, the cost is the commission times how often the bet wins, so it changes with the number you pick. The same $1 is charged anywhere from $20 to $39, which means the dollar you expect to lose is identical across that whole range and only the percentage moves. At $39 it is at its lowest, and one dollar more restarts it at the top.

Buying versus placing, at $20

  • The 4 or 10

    Buy1.67%

    Place6.67%

    Buying costs less here. Buying is the cheaper of the two at $20 and up.

  • The 5 or 9

    Buy2.00%

    Place4.00%

    Buying costs less here. Buying is the cheaper of the two at $20 and up.

  • The 6 or 8

    Buy2.27%

    Place1.52%

    Placing costs less here. Buying is the cheaper of the two at $31 to $39, level at $30.

Buying beats placing on the 4 and 10 by a wide margin, which is well known and true at any size. Carrying that to the 6 and 8 is where it goes wrong: there the place bet is already close to fair, so a $20 buy costs more than placing it, and the buy only wins back a narrow window before the next dollar of vig puts it behind again. This comparison uses the commission-on-win figures, since a vig taken up front loses to the place bet on every number at every size.

Which convention a table uses is not printed anywhere and it changes the price of the same bet by nearly three times, so it is worth asking the dealer before you bet rather than assuming. Both are in live use. A cheaper bet is still a losing bet, and putting more money down to move the percentage means more of your money is exposed to it.

Every common bet, per bet and per roll

The bold figure is the house edge per bet resolved, which is what a published craps table prints. The line under a bet’s name is what it costs per roll of the dice it is in action for, which is what you are actually exposed to at the table. A bet with no second line settles in a single roll, so for that one the two figures are the same number.

The two don’t rows are the exception to that first sentence, and it is worth a moment, because it moves the comparison the don’t side is usually sold on. A barred 12 on the come-out pushes rather than settling, so those two are priced here per bet made, pushes left under the line, which is the 1.36% a published craps table prints. Counted over only the come-outs the bet does settle on, it is 1.40%, which is the figure most other sources quote. One wager, two denominators, and neither is wrong. Against the pass line’s 1.41%, which has no pushes and so reads the same either way, the don’t side is ahead by about a twentieth of a point counted per bet made and by about a hundredth counted per settlement. Both comparisons are sound. Which of them describes your night depends on whether you are counting what you put up or what got decided, and the familiar version of the claim is the first one.

Line bets

  • Don't pass0.40% per roll, over 3.4 rolls1:11.36%
  • Don't come0.40% per roll, over 3.4 rolls1:11.36%
  • Pass line0.42% per roll, over 3.4 rolls1:11.41%
  • Come0.42% per roll, over 3.4 rolls1:11.41%

Place bets

  • Place 6 or 80.46% per roll, over 3.3 rolls7:61.52%
  • Place 5 or 91.11% per roll, over 3.6 rolls7:54.00%
  • Place 4 or 101.67% per roll, over 4.0 rolls9:56.67%
  • Big 6 or 82.78% per roll, over 3.3 rolls1:19.09%

Field

  • Field (12 pays 3:1)1:1 / 2:1 / 3:12.78%

One-roll & hardways

  • Hard 6 or 82.78% per roll, over 3.3 rolls9:19.09%
  • Any craps7:111.11%
  • Hard 4 or 102.78% per roll, over 4.0 rolls7:111.11%
  • 3 or 11 (single roll)15:111.11%
  • 2 or 12 (single roll)30:113.89%
  • Any seven4:116.67%

What the per-roll figure changes

Placing the 4 or 10 prints 6.67% against 2.78% for the field, so on the published figure the field reads as the cheaper of the two. Per roll it is the other way round: the place bet sits there for 4.0 rolls and costs 1.67% of each one, while the field settles every roll and costs 2.78%. Ranking those two on the number a craps table prints picks the more expensive bet.

Field (12 pays 3:1), Big 6 or 8, Hard 6 or 8 and Hard 4 or 10 span 2.78% to 11.11% on the published figure and cost exactly the same per roll: one unit per 36 rolls, 2.78%, on every one of them. Moving between them changes how often you get paid and how large the payout is, not what a roll of the dice costs you.

Line bets with odds are the only way to play craps at close to a fair price. Every bet in the center of the layout costs several times as much per roll of the dice, and the per-roll figure is the one that scales with how long you stand there. A cheaper bet is still a bet with a house edge on it, and none of these numbers turns into a win rate.

Written and reviewed by Alexander Korsager · last reviewed October 8, 2026 · corrections

The idea

What this tool shows

Craps has one of the widest spreads of house edge of any casino game. The line bets (pass, don't pass, come, don't come) are among the best bets on the floor at around 1.4%. The proposition bets in the middle of the layout are among the worst, running from 9% to nearly 17%. Same table, same dice, more than a tenfold difference in cost.

  1. This tool derives every figure from the 36 equally likely outcomes of two dice rather than quoting them.The pass line wins 244 times in 495, for a house edge of 1.41%; any seven pays 4:1 on a 1-in-6 shot, for 16.67%; a hard 8 pays 9:1 when it is really 10:1 against, for 9.09%. Nothing here is a house secret, it is just arithmetic most players never do.

  2. The odds bet is the one genuinely fair wager in the building: once a point is set you can back your line bet at true odds with zero house edge.It cannot make craps a winning game, but it lowers the edge on your total action. Backing the pass line with 3-4-5x odds takes the blended edge from 1.41% down to 0.37%; 10x odds takes it to 0.18%. The calculator shows the combined number for each multiple, on both sides of the line.

  3. The don't side is where that figure needs reading carefully, and it is the reason this page prices what each bet costs to hold rather than only what it costs as a percentage. Laying odds means risking more than you stand to win, because once a point is set the don't bettor is the favorite. So the blended percentage falls faster on that side, and it falls because the money on the table grew. At a 3-4-5x table a $10 don't pass laid to the maximum blends to 0.27% against the pass line's 0.37%, with $70 out on every point to win $30 to $50, where the pass bettor has $40 to $60 out to win $60. The lower number is the smaller share of the bigger stake, and the calculator prints both so the comparison is a real one.

  4. Every one of those percentages is priced per bet rather than per roll, which is the figure a craps table prints and is not the figure you are exposed to.A pass line bet takes about 3.4 rolls to settle and the field settles on every roll, so the same percentage buys very different amounts of exposure. The per-roll cost is printed under each multi-roll bet in the list, and the two rankings disagree in places worth knowing about. Placing the 4 or 10 reads at 6.67% against a field bet's 2.78%, and is the cheaper of the two per roll: 1.67% against 2.78%. On a table where the 12 pays 3:1, the field, the Big 6 or 8 and both hardways span 2.78% to 11.11% on the printed figure and cost exactly one unit per 36 rolls, all four of them.

  5. None of that answers the question a player actually asks, which is not about one bet in isolation.A real setup is a line bet with odds plus a place or two working behind it, and those percentages cannot be added together because each bet resolves at a different rate. The setup panel further down blends whichever bets you turn on into one figure and prices it per hour, at a rolls-per-hour figure sourced from a published casino-operations reference rather than assumed, because the pace of play is the one input in that calculation nobody else states.

FAQ

Questions players ask

What is the best bet in craps?

Don't pass or don't come at 1.36%, marginally better than pass/come at 1.41%. That is the per-bet-made comparison, which is the one every craps table prints. Counted per settlement instead, the two are 1.40% and 1.41% and the gap nearly closes, so the don't side is a small improvement rather than a decisive one either way. Backing either with maximum odds lowers the blended edge on your total stake further: at a 3-4-5x table the pass line blends to about 0.37% and the don't pass to about 0.27%. The calculator prices both sides, and it prints what each costs to hold, because the don't side's lower percentage is bought with more money on the table. On a $10 line bet at 3-4-5x the pass bettor has $40 to $60 out once a point is set and the don't bettor has $70 on every point, to win less.

How much can you lay behind a don't pass, and what does it cost?

Once a point is set you can lay odds behind a don't pass the same way a pass bettor takes them, at true odds and with no commission. Because you are the favorite at that stage you have to lay more than you stand to win: 2 to 1 against the 4 or 10, 3 to 2 against the 5 or 9, 6 to 5 against the 6 or 8. What a table's advertised multiple means on that side is a house rule and the two readings are not close. Wizard of Odds states one of them as a maximum on the amount won ("you may bet up to an amount such that a win would be no more than 5X the Don't Pass bet"); the widely documented 3-4-5x rule states the other as a maximum on the amount laid, at 6x the flat bet on any point. At 3x those two readings give 0.34% and 0.45% from the same table sign, so the calculator asks which one your table uses rather than picking for you. The 3-4-5x case needs no choice: laying 6x on any point is the same wager as laying to win 3x, 4x or 5x by point, so both readings land on one bet and one figure.

Is laying odds on the don't pass the same as a lay bet?

No, and the two sit on the same page here, so it is worth separating them. A lay bet is a standalone wager against a number that pays true odds less a 5% commission, and that commission is the entire house edge on it. Odds laid behind a don't pass carry no commission at all. They are the same fair wager the pass side gets, taken from the other end, and the house takes nothing on them. If a dealer asks you for a vig on odds behind the line, that is not the bet you asked for.

Why do other sites say the don't pass is 1.40% when this page says 1.36%?

Because the house bars one come-out total on the don't side, the 12 on most tables, and a barred total pushes rather than winning. The two figures put that push in different places. Count every come-out you put money up on, pushes included, and the don't pass is 1.36%. Count only the 35 in 36 come-outs that actually settle, and it is 1.40%. It is the same bet and the same arithmetic, divided by two different denominators, and neither site is wrong. This page prints 1.36% in the table because that is the convention a craps table prints, and prints the 1.40% beside it so the comparison with any other source is checkable rather than confusing. A table that bars the 2 instead of the 12 gives exactly the same figures.

Does taking odds lower the house edge on the pass line itself?

No. The pass line is still 1.41% and the odds bet is still 0%. What changes is the average edge across your combined wager, because you now have more money in action at a fair price. Per dollar risked, you lose less.

Why are the proposition bets so bad?

They pay less than the true odds by a wide margin and they resolve every roll, so the edge compounds fast. Any seven at 16.67% means that for every 100 units bet on it you lose about 17, roll after roll.

What is the difference between house edge per bet and per roll?

Per bet resolved is the average loss each time the bet settles, which is what a published craps table prints. Per roll divides that by how many rolls the bet is in action for, so it prices time at the table rather than settlements. A one-roll bet like the field or any seven has the same number for both. A pass line bet takes about 3.4 rolls to settle and a place bet on the 4 takes four, so their per-roll cost is a quarter to a third of their printed figure. Comparing a one-roll bet with a multi-roll one on the printed figure alone overstates how close they are.

Is the field bet cheaper than placing the 4 or 10?

On the figure a table prints, yes: the field is 2.78% with the 12 paying 3:1, against 6.67% to place the 4 or 10. Per roll of the dice it is the other way round. The place bet is in action for four rolls at a time and costs 1.67% of each, and the field settles every roll at 2.78%. If the house pays only 2:1 on the 12, the field doubles to 5.56% on both measures.

How much does craps actually cost per hour?

It depends on your setup and the table's pace, and the two multiply together rather than add. A $10 pass line with 3-4-5x odds costs about 0.37% of your action alone, which sounds close to nothing, and by itself it is: about $5.66 an hour at a typical full table's 135 rolls (7 players, sourced below). Add $30 working on the 6, a common real setup, and the blended figure rises to about 0.89% of a much larger total wager, roughly $24 an hour, because the place bet's 1.5% edge on its own $30 outweighs the odds bet's near-zero cost on a bigger stack of money. The setup panel above computes this for whatever you're actually playing rather than making you add percentages that do not combine linearly. A slower table lowers the dollar figure without changing the percentage; a faster one raises it the same way.

Methodology

How this is calculated

Every figure comes from the 36 equally likely outcomes of two dice, not from a published table.

A buy or lay bet is priced separately because it pays true odds, so the fair part cancels exactly and the entire house edge is the 5% commission: the expected loss is the commission times the probability the bet wins when the house collects it only on a win, and the whole commission when the house takes it up front.

Written out, that is EV = -p x v for the first and EV = -v for the second, where p is how often the bet wins and v is the commission in dollars.

Because the casino rounds the 5% down to the whole dollar, the same commission is charged across a $20 band of bet sizes, so the edge falls as the bet grows inside a band and jumps back up at every step.

Common misreadings7
  • A lower per-roll figure is not a smaller loss if you bet more often to get it. Four bets that cost the same per roll cost the same per roll whichever one you pick, and picking the one that settles fastest simply resolves your money sooner. Nothing on this page identifies a bet that makes money.
  • A lower percentage on a larger bet is not a cheaper night. The dollar you expect to lose on a buy bet does not fall as the bet grows, it stays flat across the whole commission band while the percentage falls, so the smaller number is bought with more money at risk.
  • Buying beating placing on the 4 and 10 does not carry to the 6 and 8. At a $20 buy the 6 and 8 cost more bought than placed, and the window where buying wins there is narrow.
  • The odds bet carrying no house edge does not make craps a fair game. It lowers the average cost across a larger total wager; it does not lower the cost of the line bet, and it puts more money in action.
  • The don't side's blended figure being lower than the pass side's at the same advertised multiple is not a cheaper bet. It is a smaller share of a larger number. At a 3-4-5x table a $10 don't pass backed to the maximum has $70 on the table on every point, to win $30 to $50; a $10 pass line has $40 to $60 out, to win $60. The percentage falls because the denominator grew, and the dollars you expect to lose on the flat bet do not move at all.
  • Picking the side of the line on the blended percentage is picking on the wrong number. The two flat bets are 1.41% and 1.36%, and everything below that on either side is the odds bet averaging a fixed cost over more of your money. Nothing here identifies a bet that wins.
  • A lower blended edge on your setup is not a cheaper night if it comes from adding more action rather than fewer losing bets. Odds behind the line lower the blended percentage because they are fair, not because they lower what you are staking; a place or buy bet added to the mix raises both the dollar loss and the percentage, because it has a real house edge behind it. The setup panel's per-hour dollar figure is the number to read, not the percentage alone.
Assumptions15
  • The commission is 5% of the amount bought (or, on a lay, the amount you stand to win), rounded down to the nearest whole dollar. That is the standard practice and it is the step that makes the edge move with bet size.
  • Buy and lay bets pay true odds before the commission. Place bets pay short of true odds (9:5, 7:5 and 7:6) and carry no commission.
  • House edges are a share of the money the player has at risk. For every bet here except the don't pass and the don't come, that is the same thing as per bet resolved, because the bet either wins or loses. Where the commission is taken up front, the commission is counted as money at risk, because it is.
  • The don't pass and don't come are the exception, and the exception is the reason those two are quoted at two different figures across the internet rather than one. The house bars a come-out total on that side, 12 here, and a barred total pushes: it is neither a win nor a loss. Counted per bet made, with the pushes left under the line, the don't pass is 1.36%, which is what a published craps table prints and what this page prints in bold. Counted over only the 35 in 36 come-outs the bet actually settles on, it is 1.40%, which is what most other sources print. One wager, two denominators. The page now prints both rather than picking one, and the paragraph under the table says which is which.
  • A table that bars the 2 instead of the 12 gives identical figures, because each total is one of the 36 ways. Barring anything else would not, and this tool does not model it.
  • The per-roll figure is the per-bet-resolved edge divided by the average number of rolls the bet is in action for. A bet that can settle on w of the 36 two-dice outcomes settles once every 36/w of its live rolls: 4 rolls for anything riding on the 4 or 10, 36/11 for the 6 or 8, 1 for a one-roll bet. A line bet uses the full expected length of a decision, about 3.376 rolls, because it is working from the come-out.
  • The field row's payout column reads as the ordinary 1:1 numbers, then the 2, then the 12, and follows the 12-pays control above it.
  • The odds bet is exactly fair on both sides of the line and carries no commission. Taking odds on the point pays 6/w to 1, where w is the ways to roll the point; laying against it pays w/6 to 1, which is the same fraction inverted because the don't bettor is the favorite once a point is set. Neither is the lay bet priced further down the page, which is a different wager against a 5% commission.
  • What a table's advertised odds multiple means on the don't side is a house rule, so it is a control rather than an assumption. Under the win-capped reading the multiple caps what the odds bet can return, and the amount laid works back from the point: laying to win 2x costs 4x on the 4 or 10, 3x on the 5 or 9 and 2.4x on the 6 or 8. Under the risk-capped reading the multiple caps what is laid, flat across every point, and the win works forward from it. Both are in live use and they are not close.
  • The 3-4-5x option is a documented combination rule rather than a reading of a multiple, and it takes no convention: the don't bettor lays 6x on any point, which wins 3x on the 4 or 10, 4x on the 5 or 9 and 5x on the 6 or 8. Laying 6x flat and laying to win 3x, 4x or 5x by point are the same wager, which is the mirror of the pass side, where 3x, 4x and 5x all win 6x.
  • The don't side's blended figures are built on the 1.36% per-bet-made flat edge, with the barred come-out left in the denominator. That is the basis the table on this page prints in bold and the basis Wizard of Odds combines with laying odds to publish its own don't-side figure, so the two are checkable against each other.
  • The money column is priced against a $10 line bet. The ratio between the two sides' exposure is the same at every stake, so the dollars scale with your own bet.
  • The setup panel blends several bets by putting each on a common footing, one roll of the dice, before summing. A bet's own edge is priced per bet resolved, and different bets resolve at different rates, so the edges cannot be averaged directly: each bet's expected dollar loss and its money at risk are divided by the average rolls it takes that bet to resolve, the results are summed across every active leg, and the total is scaled by the rolls-per-hour figure below. The blended edge printed is that total loss as a share of the setup's own total action, which reproduces a single bet's own edge exactly when only one bet is active.
  • The odds bet behind a line bet is exactly fair, so adding odds to a setup changes the money at risk without changing the expected dollar loss on the flat bet itself. The blended figure reflects that: more odds spreads the same expected loss over more action, which is the same effect the blended-edge percentage above the money list already shows for the line bet alone.
  • Rolls per hour, by number of players at the table, are sourced from Jim Kilby's Casino Operations Management, as published in Wizard of Odds' Ask the Wizard #136 (https://wizardofodds.com/ask-the-wizard/136/, retrieved 2026-09-16). The panel offers only the six player counts that source measured (1, 3, 5, 7, 9 and 11) rather than an interpolated curve between them, because a number this page did not measure is not one it should assert as precise.
Limits12
  • The per-roll figure counts only rolls the bet is live for. Place and hardway bets are off on the come-out unless you call them working, and a roll the bet is off for is one it has no money exposed on, so it sits in neither side of the ratio. That makes the figure the cost of a roll you are actually exposed on, and it is not a cost per roll of the session.
  • Per roll and per bet resolved answer different questions and neither one is the correct measure on its own. Per roll prices standing at the table; per bet resolved prices each settlement, which is the one to use if you are comparing what a fixed number of bets will cost. The tool prints both rather than picking one.
  • Which commission convention a table uses is a house rule, is not printed on the layout, and is not something this tool can know. Both are in live use and they are nearly three times apart on the 4 and 10, so the figure that applies to you depends on an answer you have to ask for.
  • Table minimums are ignored. Many houses will not book a buy small enough for the rounded commission to reach zero, and some set their own minimum on buy and lay bets specifically.
  • Some houses charge the lay commission on the amount risked rather than the amount won, which prices it differently again. The lay rows here use the more common convention, charging on the win.
  • The odds-multiple figures blend the flat line bet with an odds bet placed every time a point is set, at the table maximum. A player who does not back every point, or who backs it for less than the maximum, is somewhere between the blended number and the flat one.
  • Which of the two lay conventions a table uses is not printed on the layout and is not something this tool can know. Only two are offered here, because those are the two that can be pinned to a published source. If your table states its don't-side maximum some third way, the figures here do not describe it.
  • Table maximums are ignored on both sides. A house that advertises 100x odds will still have a ceiling in dollars, and the 100x column is included because tables offering it exist rather than because laying 100x is a thing many players will be booked for.
  • Bet increments are ignored. A dealer will ask for an odds bet in an amount that pays evenly, so laying $20 against the 5 to win $13.33 is arithmetic rather than a bet you will be booked for. The figures are exact at the amounts shown and a real table rounds you to the nearest workable one.
  • The don't side is ahead of the pass line on both bases, and by how much depends on which one you count. The pass line has no pushes, so it reads 1.41% either way. Against it, the don't pass is ahead by about a twentieth of a point per bet made and by about a hundredth of a point per settlement. Both comparisons are internally consistent and neither is a trick: the familiar 1.36% against 1.41% is the per-bet-made pair. What is worth knowing is that per settlement the gap is small enough that the don't side is close to a coin-flip preference rather than the clear improvement it is usually presented as.
  • The setup panel prices each point pair once, exactly like the buy and lay panel above it: '6 or 8' is one number's worth of exposure, not both simultaneously, because 6 and 8 (and the other two pairs) cost the same to bet by symmetry. Working both numbers of a pair at once, a common real setup, costs twice what that row shows, and the panel says so beside the control.
  • The rolls-per-hour source gives no detail on shuffle speed, dealer count or a private versus public table, only player count, and its own text carries no stated methodology beyond attributing the figures to Kilby's book. Treat the figure as a reasonable published estimate for a typical casino table, not a measurement of any specific one.
Evidence

Every probability is computed from the ways to roll each total with two dice, in math.ts, and checked in math.test.ts against an independent cash ledger that follows the money in and out rather than restating the same formula.

The two conventions can be checked against published tables that use them: the Wizard of Odds craps appendix prices buy bets at 1.67%, 2.00% and 2.27% (commission on the win) and lay bets at 2.44%, 3.23% and 4.00% (commission up front), and the tests reproduce all six from one model.

The per-roll column is checked the same way, against Wizard of Odds's published per-roll figures (pass 0.42%, don't pass 0.40%, place 6 or 8 0.46%, place 5 or 9 1.11%, place 4 or 10 1.67%, Big 6 or 8 and both hardways 2.78%), which are derived from expected loss per roll rather than by dividing the numbers this tool already computes.

The two don't-pass bases are checked against the same appendix, which gives 1.364% counting ties as money bet and notes that most other sources on craps claim 1.403%, the difference being whether the barred come-out stays in the denominator; both figures are pinned in math.test.ts from a ledger over the 36 come-outs rather than from either formula, so a test written in the shape of one basis cannot quietly confirm the other.

The don't side with odds is checked the same way and against the same appendix, which works the bet through at double odds: with a $10 don't pass the player lays $40 to win $20 on the 4 and 10, $30 to win $20 on the 5 and 9 and $24 to win $20 on the 6 and 8, for a combined 0.455%.

This tool reproduces all three legs and the figure from one model, and its 3-4-5x don't-side result of 0.27% matches the combined figure published by Controlled Craps for the same rule.

The two lay conventions are sourced rather than reasoned out: Wizard of Odds states the win-capped one on its craps basics page ("you may bet up to an amount such that a win would be no more than 5X the Don't Pass bet"), and the same page and Parlay Nation both state the 3-4-5x maximum as a cap on the amount laid, at six times the flat bet on any point.

Every blended figure on both sides is additionally checked against an independent cash ledger that walks the 36 come-outs and follows the money in and out, so a test cannot confirm the division without also confirming the wager, the payout ratio and the barred come-out behind it.

The setup panel's blend is checked three ways in math.test.ts: blending a single line leg alone reproduces lineWithOdds's own combinedEdge exactly, so the combination step cannot silently change a figure already checked above; a three-bet setup (a pass line, a place bet and a buy bet) is checked against an independent per-hour ledger built directly from pointBeforeSeven, placeEdge and vigCharged rather than from the blending function's own code; and adding a losing bet to a setup is asserted to always raise the total, never lower it.

The rolls-per-hour table is a citation, not a calculation: read directly from Wizard of Odds' Ask the Wizard #136 (https://wizardofodds.com/ask-the-wizard/136/, retrieved 2026-09-16), which attributes it to Jim Kilby's Casino Operations Management, and pinned as a literal table in math.test.ts so a future edit cannot silently alter a sourced figure.

Gambling can cause serious harm. The National Problem Gambling Helpline is free, confidential and open 24/7, call or text 1-800-MY-RESET (1-800-697-3738), or chat at ncpgambling.org/chat.

Read up

Go deeper

Related tools

We’d like to use privacy-friendly analytics (Google Analytics, Vercel Analytics and Speed Insights) to see which tools and pages are actually useful, so we can improve them. Nothing is sold, and nothing loads unless you say yes. See our privacy policy for details.