James ‘SplitSuit’ Sweeney On GTO Poker: Why One Bet Size Doesn’t Fit All

James ‘SplitSuit’ Sweeney has been playing and coaching poker for more than a decade, having found the game while in college. As a co-founder of the poker training site Red Chip Poker, the Chicago native is one of the sharpest in-depth strategy contributors in the industry with hundreds of videos and one-on-one coaching with more than 500 students. Now living in Florida, he primarily plays live cash games as showcased in his popular You Tube vlog, The Poker Bank.

His Game Theory Optimal (GTO) exploration began in 2016, although his goal has always been to ride the line between solver outputs and real-time exploitation. Sweeney is the author of numerous poker books, including Dynamic Full Ring Poker: A Practical Guide To Crushing $1-$2, Optimizing Ace King: Poker’s Most Complex Starting Hand, and Unfolding Poker: Advanced Answers To The Most Frequently Asked Questions.

His latest book, GTO Poker Gems: 12 Insights From The Solver That Every Player Should Use, distills macro game theory ideas into usable tips that help players both make stronger decisions at the table and study better away from it.

_Card Player _caught up with Sweeney to discuss one aspect of poker that many amateur players get wrong, which is bet sizing.

Craig Tapscott: In your new book you mentioned in chapter 5, aptly named Multiple Bet Sizes Are Mandatory, that a single-size solution doesn’t work. Can you break this down in simple terms for us?

James Sweeney: Over the years, players sort of landed on the idea of using a single-bet size for their entire range. At first glance, this makes a lot of sense. It wouldn’t make much sense for a player to bet all of their strong hands for a large size and bet all of their marginal hands for a small size. Any opponent with half a pulse would pulverize that strategy once it’s discovered.

The next logical idea would then seem to be to use a single-bet size for all hands in their range. Whether they have the nuts, marginal strength, or air, they will bet them all for the same size. This seems like a great idea since this player isn’t giving away any bet-sizing tells. 

However, there is a balance issue here. Take a simple example with a heads-up pot on the river. The aggressor decides to bet for $50 into a $100 pot with a perfectly polarized range, so 50 percent of their betting range is air and the other 50 percent is pure value. 

The aggressor is implementing a single-size solution, and while the defender has no clue whether the aggressor has air or value this time, it does not mean that the aggressor is properly balanced.

Since the defender is getting 3:1 pot odds, a properly balanced betting ratio for the aggressor would be one bluff combo for every three value combos. But since the aggressor is actually betting one bluff combo for every one value combo, the defender can correctly bluff-catch every single time to capitalize on the fact the aggressor is bluffing too frequently.

So, while a single-size solution initially seemed to be balanced and insulated from giving away bet-sizing tells, you see why it doesn’t really work.

Craig Tapscott: It sounds like it could be possible to balance with a single size though? Couldn’t the aggressor bluff less often to get a correct bluff-to-value ratio?

James Sweeney: Fair point. In that example, the aggressor could scale back their bluffs and instead of bluffing 50 percent of the time, they could check more bluffs and instead get the bluffs down to 25 percent of their betting range. But this can quickly create a very real issue.

Given the clear drop off in bluffs, more of the aggressor’s range will be checking the river. And since more air hands will be in that checking range, it will likely become impossible to not overfold when the defender bets the river.

So, bluffing the river less often may fix one issue (the bluff/value ratio), but it also creates a number of additional issues with the low river betting frequency, and higher fold-versus river-bet frequency.

A better way to think about balance is the definition used in the book, which is “balance is a byproduct of playing every hand in our range at maximum EV (Expected Value) against a perfect GTO opponent.”

Craig Tapscott: Can you give an example of how using multiple bet sizes creates that balance?

James Sweeney: Sure. Let’s build upon the situation from earlier. Again, a heads-up pot on the river, but now the out-of-position aggressor has two types of value hands in their range. They have the pure nuts, and they have thin-value hands.

The pure nuts will, of course, never lose when the defender calls, but the thin-value hands will lose to some of the bluff catchers in the defender’s calling range.

A lot of players would look at this spot and think, “Well, it makes sense to bet large with the nuts to maximize EV. And it makes sense to bet smaller with the thin value hands to minimize loss when behind and make capturing value easier against the defender’s bluff catchers.”

Given those conflicting ideas (big bets with big hands, and small bets with small hands), a single-size strategy won’t work. Indicating, once again, that multiple sizing is mandatory for a balanced strategy.

Craig Tapscott: How would a solver analyze a situation with multiple sizes?

James Sweeney: Good question. As with anything involving a solver, the solver can only function within the parameters of your game tree. Which means you need to build multiple sizes into the game tree before hitting the “solve” button. For most basic analysis, two or three sizes at each major inflection point is fine. You could certainly explore four or more sizing options, but usually there are diminishing returns once you add too many sizing options and start exploring tangents in the tree.

Once you’ve built your tree and given the solver a chance to crunch the data, you need to analyze the output.

A few notes about this to remember. 1 – A solver won’t necessarily use all bet sizing options. 2 – If a solver does utilize multiple bet sizing options, it does not mean the size it uses more frequently is any better than sizes used less frequently. 3 – Explore the strongest combos in each sizing option.

Point two is really important and oftentimes overlooked. If a solver suggests a mix, it means that the EV of all suggested hands are equal. It’s common to see a solver suggest betting a hand for size A 98 percent of the time and size B 2 percent of the time and a player walks away saying that size A is clearly best, and that size B should be totally ignored. This is a simplification method, but not pure GTO.

Anyway, let’s continue with the example from earlier. Again, a heads-up pot on the river and the out-of-position aggressor has nuttish hands, varying value hands, and bluffs. The defender has varying bluff catchers. And the board is A 7 3 2 2 with $100 in the middle.

If we give the solver two different bet sizing options, a half-pot $50 bet and a $150 overbet, we see the solver overbets 22 percent of the time and uses the half-pot bet 31 percent of the time.

While the frequencies are interesting enough, there is real value in digging into the actual breakdown of selected hands. Here are some highlights from this exploration:
There are some pure strategies used. The solver suggests that the full houses and quads (7-7, 3-3, and 2-2) purely use the $150 overbet. This is in line with the original idea of “big bets with big hands” from earlier for what it’s worth.

Another pure strategy suggestion is with the weak A-x hands like A-9, A-8, and A-6. The solver suggests that all of these combos use the smaller half-pot sizing. Interesting that this is also in line with the original idea of “smaller bets with smaller hands” from earlier too.

Another interesting pure strategy suggestion is with A-A. The solver suggests that A-A uses a pure small bet size, even though A-A can never lose given the defender’s range. While it’s true that smaller bet sizing, along with checks, potentially need a swath of stronger combos too for range protection, this A-A suggestion is largely based upon blockers.

As for mixed strategy suggestions, the solver mixes the strong top pair combos of A-K across both the small and overbet lines. When you drill down into the analysis, you see that A-K sits in a unique spot between both player’s ranges. A-K beats some of the defender’s stronger holdings (like A-Q and A-J), but loses to some of the defender’s hands, like A-7. This puts A-K in the middle ground since it’s strong, but not invincible, and as such it’s sensibly mixed by the solver.

A-2 combos are also mixed between sizes. While these hands differ from A-K since they cannot be beaten, there are blocker effects similar to the A-A combos.

Air hands are also mixed since the solver requires air across both sizes to ensure proper bluff/value ratios. The solver favors the 65 combos for bluffing (again, mixed across both sizes), which is not at all random.

Craig Tapscott: Wow. Sensory overload. (laughs) This is fascinating, but how can players use this information? This all seems theoretical to the point of having little applicable value at first glance.

James Sweeney: It’s important when getting into GTO exploration to constantly zoom in and out. Zoom into the solver output and aim to understand why certain combos get prioritized, then zoom out to understand the big-picture ideas.

In real-time, you are never pulling out a solver and cherry picking the answer. In real-time, you are better served to have principle ideas on what the solver suggests and then decide how close to solver land you should be.

For instance, in the example we just looked at, say you were the aggressor and knew that your opponent would never fold a bluff catcher on the river. Maybe they’re a fish who cannot fold a pair, or maybe they’re a player who just won’t believe you and will over-defend with bluff catchers. Either way, you could easily exploit them by removing all bluffs from your betting range and over betting all of your strongest hands.

You knew what the solver land suggestion was, you knew how your opponent would deviate from it, and you knew how to capitalize by deviating in a way that maximized EV against them. That’s good poker.

The big-picture ideas around solvers and bet sizing could be distilled down to the following:

It’s impossible to play perfect GTO poker by only using a single bet sizing strategy. The solver demands at least two sizing options in the game tree.

Strong hands tend to be in the larger betting options, but they also need to be used sometimes in small bet sizes for range protection purposes.

Bluffs need to be included in all bet sizing options, as the solver uses appropriate bluff/value ratios.

Your primary objective is to exploit your opponents as much as possible. Against a perfect GTO opponent, the only way to do this is to adhere rigidly to solver output. Against humans, there are other exploitative options available that break away from solver land (sometimes a little, often times by a large margin).

So, when looking for application, use this advice when doing your own solves, for sure. When playing, start by considering the value of using multiple sizes in a river situation first. Then roll back to earlier streets as you develop more confidence and knowledge through study.

You can learn more from Sweeney at splitsuit.com or at redchippoker.com, and find him on Twitter @SplitSuit. ♠