Small Stakes Shorthanded NL Poker Forums

Listen to Improva, he knows what he's talking about and explained it very well. Chances of almost every good reg at mid to high stakes is wrong is pretty small. Almost every good winning player 3 bets more than the "optimal" 8.7% vs 25% open even OOP.

I'm not trying to say that I 3bet around 8.7% against a 25% opening range IP, or even that we should. I definitely 3bet more than this because I feel like most villains make a lot of mistakes OOP and we should take advantage of this. I'm just trying to understand what the GTO frequencies are so I can make better adjustments.

how the hell is minimizing opponent's profit not the same as maximizing our own profit? Poker is all the sudden not a 0 sum game anymore but where we can maximize our profit and maximize opponent's profit so everyone is +EV. I'd love to play that game money just comes out of thin air. What's with this author's definition of optimal play? Dictionary definition for optimal is "most desirable". Apparently winning the most amount of money at poker and "most desirable" play is not the same.

I'm pretty sure that's what he says, that minimizing opponent's profits and maximizing our own is one and the same.

The strategies we end up with are the answer to the question: "How do I keep my opponents profit to a minimum?" and not "how do I maximize my profit?".

Either my english is pretty bad or this sentence means keeping opponents profits to a minimum is not the same as maximize my profit.

If optimal is to 3 bet 8.7 then those who 3 bet more are exploitable by a easy counter strategy which no one winning player has came up with yet even with all the EV calculation tools out there. Doesn't seem too likely.

"Of course, against an ideal opponent who never makes mistakes, maximizing our profit and minimizing his are one and the same. He doesn't make mistakes we can exploit, so we're left trying to minimize his gains from our mistakes. In game theory land we are often happy to break even."

Yeah you are right, but later he says above.

I guess it seems contradictory, but I think the point he's trying to make is that when we don't know how to exploit our opponents and therefore maximize our profit, then having a defensive strategy that minimizes losses is a good idea until we know how to adjust. At least that's how I interpret it.

Either my english is pretty bad or this sentence means keeping opponents profits to a minimum is not the same as maximize my profit.

If optimal is to 3 bet 8.7 then those who 3 bet more are exploitable by a easy counter strategy which no one winning player has came up with yet even with all the EV calculation tools out there. Doesn't seem too likely.

Hi guys (I'm the author of these articles).

First, clumsy choice of words by me there. I was talking about our general mindset/approach to the problem; are we trying to come up with a strategy that prints money from a leak we can see, or are we trying to protect ourselves from an opponent capable of outplaying us whenever we deviate from perfect GTO play (whatever that looks like).

At any rate, I thought I would post here to clarify what these articles are trying (and not trying) to do. I have not read everything posted here, but got the impression that zachd2323 may have presented them as something they aren't, so let's get that straight.

The articles build a model for playing the 3/4/5-bet game in an optimal fashion. We use assumptions and pot-odds math, and we end up with a "toy game" that becomes our approximation of GTO play for this scenario.

The motivation for the articles (I wrote them back in 2010, implementing theory from Matt Jandas then-ground-breaking game theory videos first published at Stoxpoker) was to show how a simple, yet rational 3/4/5-betting strategy could be built from "first principles".

The model is first and foremost an educational tool. It presents the total 3/4/5-betting strategy as made up of neat and tidy ranges, each with a purpose (value hands, flatting hands, 3-bet bluffs, 4-bet bluffs, 5-bet bluffs). We start with a set of squeaky tight opening ranges (core ranges), add assumptions, and do some math. The reader can then apply the methodology to his own ranges if he wants to.

The model can be implemented as it is, and it will get you started on the right track, but it was not meant to be the final GTO solution to 3/4/5-betting, not by a long shot. The model structures thinking by breaking the total strategy into tidy components. This is good. Some players take a cookie cutter approach and apply it blindly as it is, without adjusting to opponents. This is unfortunate, since cookie cutter thinking makes us lazy.

But the model is a good starting point for the newbie player who has opened KQo from UTG and faces a positional 3-bet and is unsure what to do. If he has studied the model, he knows that he doesn't have to defend hands like this one (unless he has a read-based reason to), he knows why, and he can fold without worry. He also learns how to think about value and bluffing ranges as dynamic entities, dependent on his and his opponent's position (the opening ranges set up the rest).

So take the model for what it is; an educational tool designed to aid structured thinking. It is not perfect, but it's a pretty good starting point. The articles are getting slightly old now, and people understand these things much better than they used to, but I believe the material still has value for new players.

Bugs

Here are a couple of weaknesses in the model:

1) We assume the raiser never flats OOP

This is not a necessary choice, but a reasonable one. However, against a player who does not adhere to this rule, our model breaks down somewhat. If he flats a lot of dominated Broadway hands OOP, it makes less sense for us to 3-bet a very polarized range, and we can begin 3-betting the best hands we would flat in the model. The standard way to exploit players who are loose and weak OOP.

2) The players don't go all the way in the adjustment/counter adjustment process

Here's one place where this comes into effect:

In the model we simply choose Axs as 5-bet bluffs (first using a certain amount of them in our 3-bet bluffing range, then bluff-shoving them over a 4-bet). The rationale is that these hands have a robust ~30% equity against most 4-bet/call ranges. The purpose of bluff-shoving some hands is to make it impossible for the raiser to increase EV by folding her weakest 4-bet value hands to a shove. So far so good.

But if the raiser knows we are specifically using Axs for this purpose, she can adjust her 4-bet/calling range to exploit this fact. Then the 3-bettor can counter-adjust by tweaking his 3-bet/5-betting ranges (bluffshoving more pairs and less Axs as a start) to exploit that adjustment, and on and on we go.

Theoretically then, after a long series of adjustments and counter-adjustments, the players will have converged on a strategy pair where neither player can increase EV by making another adjustment. The resulting ranges are probably not at all neat and tidy like in the model.

The model does not go there, it makes reasonable initial assumptions/choices and does not allow for infinite tweaking.

3) The opening ranges are static

A subtle point, but the raiser also has the opportunity to adjust against the 3-bettor by tweaking her opening ranges. In the model we bypass this problem by starting with core ranges (hands we never fold, regardless of our opponents). In practice our opening range for a particular position is dynamic and changes from table to table, and from orbit to orbit at each table.

IMHO I think that the biggest problem with Matt's videos and this article is the understanding of equity. It seems to be the root of the assumption about 50% equity and something called a value range.

As far as I remember Matt took some math that makes perfect sense on the river and started using it on earlier streets. I was not a big fan of that.

Yup, defining "3-bet value hands" as those with 50+% equity when getting all-in is a simplification for sure. Any hand we 3-bet/5-bet also gains EV from getting folds along the way, so it gets blurry.

The model unblurries things by making certain choices that allow us to define clear-cut ranges; "value hands" and "3-bet bluffs" (and a sub-set of the bluffs is defined as "5-bet bluffs")

Same problem in GTO'ish postflop modeling. Treating equity and it's distributions over streets correctly makes things very tricky (as a first approximation we can use a 1-street model for each street separately, and treat hands as nuts-or-air).

But a model doesn't have to be exact to be useful, and we can always refine it if need be. I've been tempted to do a second round with the 3/4/5-betting model, but these days I play PLO where everything is nice and blurry. ;-)

There's no such thing as balanced 4bet/fold strategy. 4 betting bluffing is either +EV in which case we should do it or its not in which case we should never do.

Since I'm here, I might as well comment on this too:

You're stating the exploitive view, but consider this: How do you play against an opponent who observes and adjusts, and who plays the exploit/counter-exploit game better than you do? And what is the best way to play against an unknown?

Playing GTO(ish) means we carry around lots of weaponry that we might not need. We carry it anyway to protect ourselves against possible attacks. For example, 4-bet bluffing does nothing for us against players who don't 3-bet bluff excessively, but it protects us from harm, should they suddenly change their ways without us noticing. And our 4-betting range is balanced, so that players trying to punish our 4-bet bluffs by shoving excessively will run into our calling hands too often to make a profit.

So GTO play is not only about what-is, it's also about what-might-be and being prepared for it. Striving to play GTO'ish against everyone is of course silly. Our goal to play exploitively against everyone, but some opponents are better than us in that department. GTO (or whatever our approximation to GTO is) is a fall-back option against these. It's also a solid default against unknowns (although personally I like to attack and prod unknowns to learn their ways).

To the guy who started this thread: Sorry for the hijacking, I'll shut up now. ;-)

To the guy who started this thread: Sorry for the hijacking, I'll shut up now. ;-)

Haha, I'm pretty sure this is my fault.

way to construct our range never changes. We put villain on a range and construct our range appropriately. If villain is better than us then there's no way we can defend and make villain 0EV even if we go GTO. Right adjustment is to play tighter so we're naturally less exploitable and not got to war with someone's who's better than us. Villain should not stay unknown for long and even vs unknown we still take our best guess at their range before constructing our own. Otherwise we're just clicking buttons.

The model of GTO play is also incorrect even in pure theoretical way as improva pointed out.

And by playing tighter, you are (probably) moving your strategy/range closer to GTO, even if you don't know or care what GTO for the situation is. You and I and everybody else have defaults that we use as fall-back options when we can't play exploitively.

For example, every time you say things like "I can't fold here, I'm at the top of my range" in a situation where your gut instinct makes it tempting to fold a strong non-nut hand, but you're really not sure, you are falling back on your approximation of GTO.

GTO is not all about exact math, but also lots of qualitative reasoning (the math is often too complicated to be practical anyway) that we all use to some degree. Phil Galfond does this a lot in his videos, which is one reason they are awesome. But I think we agree on the important and practical stuff, and that we're arguing mostly semantics and mindset here.

One last thing about the correctness of models:

All models are incorrect in one way or another. If they weren't they wouldn't be models but reality. That's fine if the model is useful, and we are aware of it's weaknesses. I pointed out some of the known weaknesses in the 3/4/5-bet model above. The simplifying rule that the raiser is not allowed to flat 3-bets OOP is the first restriction that should be lifted in a more refined model.

Anyhoo, that's enough GTO spamming from me in this thread, and again, sorry for derailing it.

Take care, guys!
Bugs

To the guy who started this thread: Sorry for the hijacking, I'll shut up now. ;-)

but that"s so interesting (i do not partcipate because of my level but i read all and i m sure im not the only one to do that

GTO is not all about exact math, but also lots of qualitative reasoning (the math is often too complicated to be practical anyway) that we all use to some degree. Phil Galfond does this a lot in his videos, which is one reason they are awesome. But I think we agree on the important and practical stuff, and that we're arguing mostly semantics and mindset here.

Ehm, GTO is only about exact math.

Ehm, GTO is only about exact math.

Yup, talking about applying GTO in an approximate fashion. ;-)

I got an idea earlier today and I PM'ed a suggestion to blah234 about continuing this discussion in another thread, if you guys are interested.

We could put one of the strategy pairs (CO open vs BTN 3-bet) from the Janda model to the test. I can play CO, using the approximate GTO strategy from the model, you guys can 3-bet OTB with a maximally exploitive strategy of your design and then we see what happens.

Some interesting questions we can ask:

- How much money can BTN make?
- Will BTN's maximally exploitive strategy leave him wide open to counter-exploitation from CO?
- If 3-betting very loose gains EV, but makes BTN very vulnerable to CO's counter-exploitation, is the reward worth the risk? That obviously depends on how much there is of both.

You could make two 3-betting strategies, tight and loose (or we could let the 8.7% GTO approximation in the model be the tight one) and see if BTN can gain a lot from 3-betting loose. If he can't, CO's defense should be robust. If he can, but only by leaving himself very open to attack from CO, we can discuss whether it makes sense for him to risk it.

I have Pokerazor running on a retired laptop somewhere, I could fire it up and do the EV calcs there. If we put a little work into it, it could become a very interesting strategy thread.