Glad that this one sparked some discussion. It's fascinating that, at least in the poll, we had people who thought we're almost always ahead and others who thought we were almost never ahead.
My view is that this hand is a pretty easy call.
The insight here is that I think people are narrowing the opponent's hand range a bit too much. There are a lot of low-probability combos he could have which collectively start to weigh pretty heavily against the chance he could have us beat.
What I actually did in working through this hand before I posted it was go through the conditional probability of every hand combination on every street, i.e. if he had exactly 55, how likely is it that he would (a) call preflop, (b) call the flop conditional upon his having called preflop; (c) call the turn conditional on his having called the flop, etc. Obviously this takes a LOT of time and requires you to make a LOT of assumptions. With that said, this process (essentially a matter of Bayesian updating) represents the Platonic ideal of how we should think about our opponents range. When I was playing poker actively, I never went through a hand in quite that much detail and I found it pretty interesting to do so. Anyway, here's about what I came up with:
Combos we lose to
Obviously, he could have 7
6
. From start to finish, that hand is the most consistent with his play. Fortunately, it's just 1 combination. It's almost impossible for him to have any other 76 combination since it did not make a straight draw until the turn.
The other hands that have us beat are 99 and KK, which is 6 combos total. But they need to be discounted because he often will have played them faster (especially KK before the flop). So call it 3 combos, although I suspect that is an overestimate. (Then again, this isn't a bad flop for slowplaying.)
That makes 4 hand combinations that beat us total.
Combos we beat
The most obvious hand that we still beat is 33. I don't really see how we can assume that the opponent would slowplay 99 but *not* 33. (If you're going to discriminate about which sets you'd slowplay, the logical one would be KK since your opponent is less likely to have a made hand in that case). Perhaps there are some players out there who would slowplay 99 but not 33, and perhaps *you* would play them that way, but we don't have anywhere near enough information to be that particular about *this* opponent's play. So discount the 33 combos by 50%, as we did 99 and KK, which works out to 1.5 discounted combinations.
Accounting for these hands only, we're good 27% of the time, which would make the call wrong (but not terrible).
However, what really swings the math is some of the other combinations the opponent *might* play this way:
55- Nobody mentioned 55 as a possibility, since you wouldn't expect an opponent to make it to the river with that hand as a default. But the opponent may be a calling station. Or he may decide he wants to continue on with a hand with some showdown value depending on how our 2-barrel/3-barrel dynamics have been working -- our hand is actually not all that well defined here. If 55 has made it to the river, it will certainly put in a big raise once it makes a set. So the 3 combos must be discounted fairly heavily, but they're probably worth 1/2 a combo or 1 combo, I think.
AA - If we're allowing for the possibility that our opponent slowplays things in general, including potentially KK before the flop, then AA needs to be considered too. More broadly, both good and bad players play AA in all sorts of weird ways, and when they play it weirdly they often overplay it. AA is a plausible hypothesis surprisingly often when the opponent's play is otherwise strange but he wants to get a lot of money in. There are 6 AA combos -- we certainly need to discount them quite heavily but call them 1 combination after doing so.
AK - While playing AK in this way does not seem very optimal with these stack sizes, it is one of the liver hands with 12 combos. Even if you discount them by 95%, that works out to 0.6 combos. Because there are so few *winning* combinations he might have, that actually makes a material amount of difference in calculating our equity.
Two Pair - This is not a very good board for two pair given his preflop range. Nor is it clear that if he had two pair, he'd play them in this way. Still, an opponent who plays Kxs before the flop -- and some opponents will even if "we" don't like that hand -- will have made two-pair with K9s, K8s, K5s and K3s. 98 is less likely because the 8's are extremely dead. Still, the two pair hands combined were worth something like 1 combo when I summed the estimates of the probabilities up.
As I said, I think it's these "exotic" hands like AA/AK/55/two-pair that really make the decision pretty clear in my view. The possibility that he'd play any *one* of these hands (e.g.
exactly A
A
) in this way is fairy low. But it's certainly not zero, given that we have a fairly uncertain read on him. I've given you fair amount of information about the opponent and we've formed some suppositions about his play, but I think we're being a bit overconfident about our read on him if we're entirely eliminating hand combinations like AA or 55.
Finally, there is the possibility of a bluff. This is more of a garbage-in, garbage-out assumption since we don't really know that much about his long-term bluffing frequency. But there are a fair number of hand combinations he could get to the river with, and if he bluffs with them even 1% or 2% or 5% of the time, they start to weigh into the equation a little bit given how few winning hand combos he has. I don't know that our hand would be worth a call if it were solely a bluff-catcher, but it certainly gives us a bit of extra wiggle room if we think the decision is otherwise close.
The moral of the story: I suspect *most* poker players are quite good at estimating these hand ranges *most* of the time. But I'm sure we also have some blind spots where our heuristics fail us, and that we tend to err on the side of narrowing the opponent's hand range too much. I know I started to have a lot of trouble in the mid-high limit games when the good players started opening up their games a lot, and the instinctual hand ranges I had developed after a couple of years of playing in the games were thrown off.
This is in accord with people's general tendency to be overconfident when they make predictions, which is a big theme of the book. It seems a lot of what people like durrrr and Galfond did was to detect and exploit cases where people's intuitive estimates of hand ranges are off-kilter.