Choosing a bet size for a stated goal

A bet size needs a job. If you are betting for value, which worse hands might call? If you are bluffing, which better hands might fold? The amount you put out changes the price, but a price alone cannot tell you how an opponent will respond. State the goal and the response assumption before deciding what the number means.

This lesson compares checking with four bets into a fixed 100-unit pot. Two fictional response profiles show how different assumptions can change the highest-return choice in the same menu. The profiles are written for the exercise. They are not observations, solver outputs or predictions about a player, poker room or stake.

Ace: “Should I use the tallest stack?” Victoria: “Tell me what you want the bet to do. Then tell me which hands you expect to continue at that price.”

Name the goal before the size

For a value bet, identify hands that can call and lose. Also account for hands that can call and beat you. A high calling frequency does not establish a good value bet if the calling range is too strong. If raises are possible, you need that branch and your intended response as well. The label “value” does not supply any of those missing facts.

For a bluff, identify better hands that can fold. A hand that sometimes wins at showdown has something to give up by betting. Checking may preserve that value without risking more chips. A bluff comparison should therefore use what checking achieves as its baseline, rather than assuming that any pot you do not bet is automatically lost.

Earlier streets add more decisions. A caller can improve, you can improve, and further bets can change the cost of reaching showdown. Here the action is heads-up on the river, with no raises, ties, rake or tournament payout effects. Both players have enough effective chips to bet and call every size in the menu. These assumptions define the exercise; they do not describe every real river spot.

Keep the pot and your new cost separate

Let P be the pot before you act and B be your proposed bet. The pot after you bet is P + B. If the opponent calls for the same amount, the final pot is P + 2B. These are three different counts. A phrase such as “the pot is 150” needs a timing label before it can be used in a sizing calculation.

With 100 in the pot and a bet of 50, the pot becomes 150 after your bet and 200 after a call. If you win, your decision-point net return is 150: collect 200 and subtract the 50 you just added. Your own bet comes back inside the award. Counting the full 200 as new profit would count those same 50 units twice.

If the opponent folds to that bet, the pot you collect includes your own uncalled 50. Your net return from the decision is the original 100. If the opponent calls and you lose, your net return is -50. Money committed on earlier streets is already inside the starting pot; this receipt is not a calculation of your profit for the entire hand or session.

Read the response profile as an assumption

The menu contains bets of 25, 50, 100 and 150 into 100, plus checking. In the profile where calls fall sharply, the assumed calling shares are 80%, 65%, 40% and 20%, in that order. The matching folding shares are 20%, 35%, 60% and 80%. Only a call or a fold can occur after a bet in this model.

In the profile where calls fall slowly, the same sizes receive calls 90%, 85%, 80% and 75% of the time. Folds are therefore 10%, 15%, 20% and 25%. The names describe these supplied rows. They do not classify an opponent as a certain type, and they give no reason to expect the same response on a different board or after a different action sequence.

There is no estimate between the four sizes. A bet of 60 is outside the menu, and a bet of 200 is outside it too. The tool cannot use these four rows to prove a best size among every legal option. It also does not decide whether a bet is legal for the actual game, action and effective stacks you are reviewing.

Value mode counts what a caller adds

Value mode makes an extreme assumption: your hand beats every opposing hand, including every caller. Checking wins the existing 100. A bet also wins those 100 whether it gets called or not, and a call adds the opponent’s bet-sized contribution to your return. Your own new chips are subtracted when the final pot is counted.

If the call probability is c, the expected net return is P + c × B. The improvement over checking is c × B. This formula works here because the existing pot is won in every branch and every caller loses to you. It is not a general value-bet formula for a hand that loses to some calls, gets raised or sometimes ties.

Under the sharply falling call profile, a 25-unit bet adds 0.80 × 25 = 20 above checking. A 50-unit bet adds 0.65 × 50 = 32.5. A 100-unit bet adds 0.40 × 100 = 40, while a 150-unit bet adds 0.20 × 150 = 30. Expected net returns are therefore 120, 132.5, 140 and 130.

The 100-unit bet has the highest return in that value menu. The 150-unit bet makes a larger pot when called, but it is called less often under the supplied assumptions. Its extra 50 on a call does not make up for the calls that disappear. A larger successful award and a larger average return are different comparisons.

Change the calls and the comparison changes

Keep the same value hand and switch to the slowly falling call profile. The four improvements over checking become 22.5, 42.5, 80 and 112.5. Add the existing 100-unit pot to get expected net returns of 122.5, 142.5, 180 and 212.5. The 150-unit bet now has the highest value among the listed actions.

Nothing about the hand or starting pot changed. The response assumption did. In the first profile, enough calls vanish to make the larger value size perform worse than the 100-unit bet. In the second profile, the larger size keeps enough calls to earn more. Neither calculation establishes how often a real opponent calls your next bet.

Ace: “The hand stayed the same. Why did the preferred size move?” Victoria: “We changed how often the model opponent calls. The result depends on that row as much as it depends on the stack we bet.”

Caller strength matters alongside caller frequency. A real larger bet may remove weaker hands while retaining hands that beat yours. The all-winning value model deliberately leaves that issue out. If you cannot justify its assumption for a hand, keep its result as an arithmetic exercise and build a review that separates winning and losing called outcomes.

Bluff mode pays for every failed attempt

Bluff mode uses the opposite extreme. Your hand loses if called and also loses when checking. Checking has a net return of zero from this decision. A fold wins the existing pot P, while a call loses your new bet B. There is no remaining card, future bluff opportunity or showdown win to rescue the called branch.

If f is the folding probability, expected net return is f × P – (1 – f) × B. The first term weights the pot you win when the bluff works. The second weights the bet you lose when it fails. With P = 100 and B = 50, a 35% fold assumption gives 35 – 32.5 = 2.5 units.

Set that expression equal to zero to find the break-even fold share: B / (P + B). Into a 100-unit pot, bets of 25, 50, 100 and 150 need 20%, one third, 50% and 60% folds respectively to break even in this zero-showdown-value model. Equality returns zero; a positive expectation requires a larger fold share.

Do not substitute the final called pot into that denominator. A bluff is risking B to win the pot that existed before your bet. For a 50-unit bluff into 100, its threshold is 50 / 150, not 50 / 200. The latter denominator belongs to a different question about the price facing a caller, not the bettor’s bluff break-even rate.

More folds can still be too expensive

Under the sharply falling call profile, bluff returns for bets of 25, 50, 100 and 150 are 0, 2.5, 20 and 50. The 150-unit bluff wins 100 on 80% of the model outcomes and loses 150 on 20%: 80 – 30 = 50. It has the highest expected net return in this particular menu, above checking at zero.

The 25-unit bluff also returns zero in that profile. It wins 100 on 20% of outcomes and loses 25 on 80%: 20 – 20 = 0. It ties checking locally, but that is not a tie for the highest return across the whole menu. A comparison must say whether it covers two chosen actions or every listed action.

Switch to the slowly falling call profile and the bluff returns are -12.5, -27.5, -60 and -87.5. Checking at zero is now the highest-return listed action. The biggest bet does get more folds than the smallest bet, but those extra folds are too few to cover the much greater loss when it is called.

A menu containing only those four bets could make -12.5 look like an attractive result because it is the least negative. Keeping checking in the receipt exposes the missing alternative. A positive improvement over a worse bet is not the same as a positive improvement over the action you could take without adding chips.

Use the desk to test a dependency

Choose a goal, a fictional response profile and your proposed action before comparing the rows. The desk lists checking and all four bets with assumed call and fold shares, expected net return, and improvement over checking. Your proposal stays visible beside the highest listed model return and the gap between them. That gap is conditional on the selected assumptions.

Begin in value mode with calls falling sharply and propose 100. Compare it with 150, then keep the proposal and change only the profile. The highest-return listed size moves to 150 while your proposed 100 remains visible. You have isolated the effect of the assumed response, rather than changed the hand and several inputs at once.

Now choose bluff mode. Under sharply falling calls, compare the 25-unit bet with checking and note their exact zero-return tie. Then switch to slowly falling calls, where checking exceeds every bet. The desk also prints the pot after your proposed bet and the final pot after a call, so you can audit the timing and your own new cost.

Changing any selection clears the old receipt. Reset restores value mode, sharply falling calls and a 100-unit proposal, then returns focus to the goal. The worked assumptions and calculations remain readable without JavaScript. The desk uses no real player identity, hand-history upload or account information, and it supplies no estimate of a real opponent’s behavior.

Keep checking value in a real review

Both modes give checking a deliberately simple value: 100 when your hand always wins, zero when it always loses. Many hands fall between those extremes. If a check would sometimes win the pot at showdown, use that opportunity when comparing a bluff. You cannot transfer the zero baseline to a hand with showdown value merely because you decided to call it a bluff.

Checking can also allow another action when you are first to act. Our fixed exercise treats the check branch as ending at showdown. A real review must include a later opponent bet if that branch is available, together with your response. Writing “check” in a table does not by itself establish that you get a free showdown.

Extend the model one missing branch at a time

For an actual hand, record the pot before your decision, effective stacks, legal options and action order. Write down what checking can lead to. Then describe responses separately for each size: which worse hands call, which better hands call, which hands fold, and which hands can raise. A single calling percentage does not tell you the quality of the caller’s range.

If called outcomes can win, lose or tie, give each its proper net receipt. A tied heads-up pot returns a different amount from a loss, and your new contribution still needs to be subtracted. If an opponent raises, include the additional cost of calling or the chips lost by bet-folding. If rake applies, state how it is charged rather than inserting an unexplained deduction.

Ace: “The worksheet prints an exact return. Can I trust the size?” Victoria: “You can check the arithmetic. The calls and folds still need evidence. Keep the assumption beside the result when you review it.”

Earlier-street sizing also needs future cards, future action and equity realization. Tournament decisions may need prize and elimination effects beyond chip return. These are material changes to the problem. The river desk does not silently account for them, and a higher fictional chip expectation does not establish the right action in a different setting.

When the response branches are uncertain, write the uncertainty down. Compare explicit alternative assumptions or identify the particular hand group whose response would change the decision. Do not assign a precise call share simply to make a worksheet complete. Sometimes the useful conclusion is that the sizing question depends on information the review does not yet support.

Finish a sizing explanation with the condition that drives it. For example, the larger value size needs enough worse hands to keep calling; the larger zero-value bluff needs enough additional folds to pay for its added risk. Those statements can be checked against a range and action history. A familiar bet fraction cannot replace that work.

Back to top