Merging GTO and Exploits
A street-by-street evolution
Imagine you’re running deep in a tournament, and you’re in a hand against a player who’s obviously not a pro. They have a reasonably clear set of tendencies, in line with what you would normally expect from someone in this player demographic, but not so obvious that it becomes a case of just running them over with aggression or never bluff-catching on any runout.
In this scenario across the entire hand, how do we decide between making the theoretically ‘correct’ play (and thereby protecting ourselves against any incorrect assumptions or potential adjustments from our opponent), and making an exploitative read that might allow us to boost our EV in this specific spot, but also risks being incorrect and resulting in an EV loss?
Is it as simple as saying “we don’t have enough info about our opponent to justify playing exploitatively, therefore we should just try to play GTO throughout the entire hand”? Or perhaps the opposite, that we should just go with our read, and trust that the upside is likely to outweigh the downside?
Well, in reality it’s neither of these things. We don’t have to lock ourselves into specific macro-level approach to playing hands — it’s possible to shift from one approach to the other over the course of a hand, and in many cases, it’s by far the best way to approach a spot.
Over the course of this post, I’ll outline why I think this is more or less the best way to approach strategy development, since it’s applicable to virtually any environment or game type, and allows the maximum scope for boosting EV through exploits without over-complicating our strategy or causing us to arrive in near-zero-frequency later-street spots with incoherent ranges.
Why merge GTO and exploits?
A valid question, especially when we consider that these two things are two sides to the same coin. It all comes down to how difficult it is to accurately predict the impact of potential later-street mistakes on earlier-street decisions in a game as complex as No-Limit Hold’em.
On the earlier streets, we’re always dealing with a scenario where there are multiple potential decisions left to be made. Even if we only take into account one combination of two hole cards on our part, there are 46 possible rivers to follow every turn spot, 47 possible turns to follow every flop spot, and 19,600 possible flops to follow every preflop spot. This leads to a final number of 42,375,200 possible board runouts for every combination of hole cards.
Even if you have a decently accurate idea of your opponent’s overall tendencies, it is near-impossible to accurately estimate the impact of their potential later-street tendencies on the most appropriate preflop or flop strategy, and yet these tendencies will significantly effect the EV of decisions on earlier streets - we just can’t necessarily accurately estimate how these effects will manifest, especially in real time.
For example, let’s say we think that our opponent is going to over-fold the river to certain bet sizes when a flush card completes. How should that impact the way we play our flush draws on earlier streets? Does it make us more inclined to grow the pot with larger bet sizes, to compensate for the fact that we can’t easily get paid off when we hit? Or does it make us more inclined to do more checking, keeping pots smaller and potentially encouraging a less significant overfold on the river? We just can’t easily figure out the answers to these questions.
Even an advanced solver like PIOSolver 3.0, which has the capacity to account for broader, less specific tendencies across all potential future streets, can’t necessarily easily account for things like over-folding some river cards while under-folding others, or being value-heavy when betting one size compared to being bluff-heavy when betting another size.
In short, it’s just virtually impossible for you, a human being, to accurately assert the impact of your opponent’s later-street tendencies on the optimal earlier-street strategy. Even if you know in great detail exactly what kinds of mistakes your opponent is likely to make on later streets, it’s not easy to figure out how that should actually alter your strategy earlier on.
Which, when it comes to strategy-building, leaves us back where we (ideally) started; an entry point of understanding the GTO or Nash equilibrium strategy for a specific earlier-street spot. You see, while it’s very difficult to estimate or accurately solve for the impact of specific later-street mistakes on the optimal earlier-street strategies, it’s actually incredibly easy to account for the impact of earlier-street mistakes on later-street strategies.
In fact, this is basically what all existing solvers already do when you implement node-locking. They solve for the locks you’ve implemented on the current street, while assuming all players continue attempting to maximize EV on the later streets, given the restrictions you’ve applied to the strategies.
In other words, if you node-lock your opponent to over-fold the flop to a specific bet size, it’s going to assume they do exactly that, but then on the turn, it’s going to recalculate the turn with the assumption that both players know that the flop has been over-folded; it’s going to establish a new equilibrium for the turn spot, which will require a new strategy from both players.
While this is far from a perfect methodology, given that anyone who makes any kind of substantial mistake on the flop is likely to make more substantial ones on the later streets, it’s a lot more of a secure foundation for a well-rounded strategy than we would have if we attempted to construct an earlier-street strategy which takes into account all of the potential weaknesses which might exist across various board textures and runouts.
Compounding weaknesses across multiple streets
One reason why it’s very difficult to make massive adaptations preflop - for example, raising to different sizes with different classes of hands, or having both a limping range and a raising range - is because any opponent is almost always going to retain some degree of board coverage in their range when going to the flop, even if they’re playing very suboptimal poker.
Even if they’re a huge whale playing any two cards, they’re still going to possess at least some strong hands on any given flop. The times where they’ll have less board coverage might be in 3-bet or 4-bet pots, where certain boards will miss their range altogether; but in those spots, the SPR is lower, and the likelihood of progressing to the river with many decisions left to be made is greatly reduced.
As a result, when we’re looking at adapting to an opponent on an earlier street, we’re unlikely to really be able to go much further than to categorize our adjustments in terms of frequencies - we might be raising a little wider preflop because they’re not 3-betting enough, or c-betting more aggressively overall because they’re not check-raising enough, for example.
It’s unlikely that we’re going to be able to confidently state in any given preflop or flop scenario that our opponent is completely capped, because they can almost always have at least some hands which can improve - even if they never flat-call preflop with a pair better than 99, they can still flop sets on most boards, and so on.
By contrast, as we move into the later streets, the mistakes they made on the earlier ones are now going to compound; we’re going to move closer and closer to a point where their range is entirely capped, or where we can have a significant degree of clarity on exactly what they’re going to be doing with different parts of their range.
This is particularly true when we’re dealing with live poker, and we have behavioral reads or tells we can add to the equation - when an opponent is making no effort to balance their physical behavior, we may reach the later streets with a very clear sense of which portion of their range we’re up against, or they may exhibit behavior on the river which gives us a lot of comfort in how we interpret their range.
In short, the closer we get to the river, the more information we have to work with - not only about the board itself (given that there are no more cards to come), but also about our opponent’s range, behavior, comfort level, and likely strategy as a result.
Studying in productive ways
While it’s extremely difficult to learn optimal poker strategy purely by memorization (and you shouldn’t rely on this as a learning technique), there are some aspects of strategy which can be memorized relatively effectively. For example, if you’re a 6-max online cash game player, it might not take very long at all to memorize your exact opening ranges and frequencies from each position.
The fact that these aspects of strategy can be learned quite quickly means they serve quite well as ‘low-hanging fruit’ when it comes to the study process. While it might take a very long time to learn how to use a preflop solver to the level of competency necessary to establish exploitative preflop opening ranges for your environment, it takes very little time to look up an optimal preflop opening range for a certain size, and learn to implement it relatively accurately.
What this means is that the earlier in the game tree we’re focused on, the more easily and rapidly we can arrive at a strategy which captures a baseline level of EV, and frees up our mental energy for studying more complex scenarios and exploring later streets.
In addition, earlier-street strategies can usually be simplified very easily. Things like eliminating flat-calling from certain positions preflop in raked environments, or simply choosing to range-bet a postflop spot instead of trying to find that 10% of hands which wants to check, are going to have very minimal impact on our EV, and will allow us to completely eliminate the need for ever studying certain lines on certain board textures. If you choose to implement that 10% checking frequency on the flop, you need to know approximately how to play the turn and river in that line, or else you’re losing EV altogether by implementing a check.
On top of all of that, if we take some kind of highly-exploitative line on an earlier street where we construct our range in a totally unconventional way - for example, betting big with all our value hands on the flop and betting small with weaker hands - then we now end up in a turn spot where we have what I often refer to as a ‘butchered’ or ‘incoherent’ range.
We have no bluffs any more, and so even if we do identify a specific runout on which our opponent might be over-folding (or if they give off a very weak behavioral tell, for example), we can’t take advantage of it, because we only have value hands! This is especially bad in spots where we’re trying to get paid off with our value - we end up with no ability to actually win big pots, because we basically have the exact type of hand our opponent expects us to have in a lot of cases.
On the river, however, we can more or less do whatever we want when it comes to exploiting, especially against opponents who will have predictable responses versus certain bet sizes. We can use some sizes that the solver very rarely uses, such as betting smaller from in position in the knowledge that our opponent is very unlikely to check-raise, we can pick some unconventional bluff sizes based on assumptions about villain’s folding frequencies, and so on.
We can do all of this in the knowledge that there are no future decisions left to come after the river. There’s no consideration of how our range is going to play on a future street, and if we’re shoving all-in on the river, there isn’t even any consideration of how we’re going to respond against a raise. It’s more difficult for any weaknesses in our opponent’s strategy to be offset by the presence of at least some strong hands in their range, because there will in fact be many spots where the compounding of their weaknesses results in their range becoming entirely capped. This can cripple their EV and boost ours far more than any earlier-street weaknesses ever could.
A sustainable strategic approach to the game
All in all, I tend to err on the side of teaching an approach to the game which is grounded in these ideas - playing somewhat close to an optimized approach on the earlier streets (although in many cases it can be unnecessary to randomize your actions), and veering more and more into heavy exploitation as we move towards the river.
When people talk in a way that implies they’re locked into a specific way of thinking - either that GTO or exploitative play is the only way to go - it tells me that they’re not really operating with a correct definition of each of those two terms. As I mentioned before, they’re two sides of the same coin.
GTO play is how we maximize EV against an unknown strategy; exploitative play is how we maximize EV against a known strategy. The earlier we are in the game tree, the more we have to account for our opponent’s strategy being at least somewhat unknown, because there are simply so many possible runouts and so many possible weaknesses which exist. If you’re telling yourself you have absolute clarity about all your opponents’ exact weaknesses across all runouts, you’re just wildly overestimating yourself.
Once we get into the later portions of the game tree, that’s where the real creativity of the game comes into play, and we should be looking to save our mental energy and strategic focus for these spots. Messing around with weird preflop strategies and ‘creative’ flop sizes is just as likely to end up leaving us in a bizarre river spot as it is to generate additional EV for us.
Which isn’t to say that there’s nothing we can do on earlier streets to improve our EV, just that it’s mostly a case of small tweaks here and there, and it’s best viewed in terms of a couple of key ideas - linearity or polarity (in terms of how our ranges are constructed), and frequencies (in terms of how much aggression we want to show in certain spots).
We might construct a more linear 3-bet range against players who are under-4-betting, or a more polar one against players who are over-folding; we might c-bet more aggressively versus players who rarely check-raise; we might even decide to go for a wider flat-calling range on the button when there are passive players or weaker players in the blinds. But each of these requires a relatively minimal amount of mental energy in order to implement, and it doesn’t alter the shape of the postflop game tree to a substantial degree.
If you’re currently in the camp of people putting a lot of energy into earlier-street adaptations and avoiding looking at more complex later-street spots, consider flipping your approach to studying on its head; it might help you open up exploits that you haven’t previously considered. In particular, it helps to balance the desire for creativity and freedom in building strategies with the ability to maintain an overall well-balanced approach, which is crucial when you’re dealing with both stronger and weaker players in your pool.
That’s about it for this week - I’ll be back next week with some more thoughts. In the meantime, GL everyone.
Matt
P.S. I’m struggling to get the RSS feed on this page working. If anyone is having trouble with it - or if anyone is actually able to view my posts through RSS, unlike me - then please do let me know. Still getting used to this platform.