Yeah I run spots all the time as you describe where you’re comparing how range is played on different runouts for different spots. Been studying that way for a long time. I appreciate being able to play the hardest games on the internet, according to some people. If I want to crush the toughest live games someday, beating these seems like a reasonable objective.

On the edit. Please point out the significant mistakes in the 4 hands I recently posted vs the “fish”. I post bad hands I’ve played all the time. Not sure any of these 4 from my earlier session fit the bad category even versus fish. And how do you know they’re fish, all the hands are posted anonymously?

How many hours a day do you play/study, no bullshit answer, actual average of productive time

First hand AK, cbet flop. It's a flop that misses you both but you have the range advantage. Once you check behind you give him hope. He leads minbet. What can he be other than a fish? I would just try to realize my equity because he will continue any pair or pair + draw and he has plenty. What are you denying equity from when raising? QJ? What are you going to do when the river bricks?

Second hand QJ why check behind the turn when you hit? I'm pretty sure that's not solver approved.

Third hand AK, another weird check behind on the turn with something that is very high in your range but not invulnerable. Maybe I'm wrong but whatever, it doesn't matter that much.

Fourth hand AQ, either going for it all the way or taking the free showdown with the little showdown value. In theory villain should block bet with his medium strength hands so you are bluffing vs hands you already beat. In practice they have a lot of check-calls.

Okay well thank you for sharing your opinions at least it makes for some interesting thoughts.

First hand, checking range seems pretty good, two bets from it going in regardless if we want it to. It's not about giving him hope. To be honest his range is probably more pair heavy than my range is even if we do have better pairs on average. Giving a free turn isn't giving up much here. I'm all for not splitting range OTF but cold call 3b cold call 4b has an inordinate amount of 88-QQ and on this particular board, 88 and 99 likely fold to the turn bet/river jam anyways. On your second point the minbet means nothing. In other spots sure, in this spot it's an exploit a lot of strong players use because the pot is so big the minbet leaves at least some % chance for accidental folds or if your opponent somehow timesout you auto win. I see this a lot once pots get huge in ACR, it's basically a check. So as played you prefer bet flop but hate this bet on turn, why? SPR is so low it only takes 2 streets to get the money in. And as pointed out this isn't raising to deny equity, this is a bet, the minbet means absolutely nothing. River bricks this hand is probably a give up, spades firing away.

QJ I mixed frequency check/bet and rolled check. Having the nut straight here is a pretty nice trap when opponent checks, since villain should never be checking turned straights when check raising this board on the flop. Therefore his range is mostly polar when checking to 3 outer, bricked bdfd's, and some 9x and 8x that decide to check now. We bet most of the rest of our value range here.



Solver agrees, so spot 3...



Mixed frequency call OTF.



Calling a lot without diamond combos.

Solver pure jams AK when checked to on the turn but this opponent is way overly bluffy so I figured w/e let's let him find a bluff on the river.

AQ spot was interesting and actually need to look up more spots like this but you're right Qd is the nut bluff card solver runs on basically all diamond runouts so I need to just be 3 barreling this combination, pretty much always.

But after looking into some of this I'm still trying to figure out where I was just donating money to fish or severely misplaying. Seems like the biggest misplay was not running the AQo hand for 3 streets, even then the solver sometimes takes it and checks turn/bets river, just not very frquently.

Yes, I was completely wrong. I didn't even see the diamond you held, which gives you the safety net. But I had some doubts (never say never) and I was thinking about this hand. I was wondering if it's the same principle of checking straights in this spot when the flush also comes. So, out of curiosity I checked it myself, unfortunately there's a 30min policy beyond which I can't edit bad advice.

The good thing is that the EV of betting is similar to checking and you can't go wrong. Another interesting observation is that the presolved solution actually prefers folding on the river. So my intuition was that since we're not folding and we want to build the pot, better bet on the turn.

Even if you’re a 2.5bb/100 winner (which would be pretty impressive), you have like a 20% chance of going broke over 100k hands

Still waiting for someone to post a 5bb 50nl blitz graph over a good sample of hands

Look I don't mean to be rude but what do people not understand? You don't have to go broke when you move up in stakes. You just have to stop if you lose a couple buy-ins. Lol.

People who can achieve this don't have an incentive to play 50b

What would be considered a good enough sample? Wouldn't mind trying this in the future but would need to have some xbook action to make it worthwhile. 5bb/100 over 100k would be around $2500 profit right?

If it's just about opportunity cost and not reputation then I feel like Doodoo and Aner0 could find a fair compromise.

What about 3bb/100? A 6bb/100 winner would have like an 83% chance of beating 50NL for at least 3bb/100.

So how come Pluribus can deduce it’s 'true' winrate over 10k but seemingly nobody else can?

Can It?

Apparently.

i wish I was that good

When you play against a solver just add up all ev mistakes that's how much you are losing to it. This converges faster then WR.

So they probably use something like that to adjust EVs.

Don't we need to know every player's full range and responses for a true winrate? As in, Pluribus' responses will often not be optimal gto against non balanced play, therefore it could be being exploited in other parts of the (unplayed) gametree that will not feature in a combo vs combo EV calc but would emerge over time in a WR, range v range.

Or are you saying it will always be close enough?

I think I read that the way they adjusted pluribus winrate was to adjust for preflop hand distribution. I didn't read into it in any detail, but my guess is that they just computer the winrate for each hand, and multiplied that by the frequency you should get each hand dealt at random. Or something like that but a little more complex. Pretty sure the only adjustment they made was to preflop hand distribution. On that basis (and I might well be wrong, I didn't do a deep dive), pluribus adjusted winrate is just more likely to be accurate than if it was the winrate as it would appear in PT4. But by no means is it an exact winrate.

However, based on the way the team described a previous HU bot (Claudico) as having achieved a "statistical tie" Vs humans (which was really stretching things, and either demonstrating a pretty poor understanding of statistics, or a keenness to twist the statistics to paint their research in a positive light which is unprofessional), I can't say I have huge confidence in their methodology. And they've clearly painted it as more accurate than it actually is.

Doubt computer scientist don't understand basic statistics. I think they just wanted above 0 WR with statistical significance ( don't know how much that is in computer science. In physics its 7 sigma).

I wouldn't rule it out, they might not have as good an understanding of statistics as people assume. There has been research showing that people who publish academic papers have a poorer understanding of statistics than they think they do. In any case, they probably have an understanding of confidence intervals.

But if they do have a sufficient understanding of statistics, then they clearly have no qualms in twisting those to paint their research in a more positive light than their results provide. Which happens, but is a little unprofessional I'd say.

I did have a quick look into the methodology they use to reduce variance, and it is actually considerably more complex than I had thought. Assuming it is correct (and I see no reason it wouldn't be), they find that their software beats a group of "elite players". I don't really know the names for many of the top 6-max players, but the only one of the list I did recognise as being a top 6-max player was Linus Loeliger. I'm not convinced that many (if any) of the others would sit at the highest stakes 6-max tables online, so are they really elite players? Linus lost to pluribus (5 pluribus and him at the table) for 0.5bb/100, with a SE of 1bb/100, so a 95% confidence interval of -2.46bb/100 to 1.46bb/100 (if I've understood how they use SE, I don't usually use those).

Basically, this has been advertised as AI having beaten the top humans at 6-max NLHE. The only valid result I could see facing a top human, it looks like the AI is more likely to have won than lost, but even with a 68% confidence level (roughly 1 sigma, and again, subject to me not misunderstanding SE), I don't think you can say it won. And with the Claudico bot, they said it was a "statistical tie" when the bot lost by over 9bb/100 over 80k hands.

Clearly, they're doing something right and their AI is pretty good, but the twisting of results to promote their research is in poor taste in my view.

This thread was too much awesome for the strat forum!

Agree, so glad I found this thread. I have already learned or (maybe clarified some foggy ideas) to apply to my own situation.

Hey all. Been a little while but have been working/traveling. Since last post have spent almost all of my time at 10NL blitz. Also had the opportunity to play some 10NL on Pokerstars recently and literally went up 7 BI in 200 hands. Back to the theme of ACR being quite difficult compared to basically any other pool.

Played an interesting hand IRL https://forumserver.twoplustwo.com/1...flame-1803029/

Otherwise just grinding/studying. I think solver approved overbet strategies are generally working a lot better than they probably should be. So I've changed up pace a bit there, leaning toward bigger bet sizes in a lot of spots as well. I'll post the chart from the past couple of weeks and some hands I played from this morning session. Nothing too crazy. Bankroll back up near $1k, so going for it again for up to 2 BI at 50NL again soon.



3b stuff, make hands:

Yatahay Network - $0.10 NL FAST (6 max) - Holdem - 6 players
Hand converted by PokerTracker 4

BTN: 160.4 BB
SB: 141.2 BB
BB: 75.5 BB
UTG: 80.3 BB
MP: 19.2 BB
Hero (CO): 206.1 BB

SB posts SB 0.5 BB, BB posts BB 1 BB


Pre Flop: (pot: 1.5 BB) Hero has 9 K

UTG raises to 2.5 BB, fold, Hero raises to 8 BB, fold, fold, BB calls 7 BB, UTG calls 5.5 BB

Flop: (24.5 BB, 3 players) 8 J 7
BB checks, UTG bets 6.1 BB, Hero calls 6.1 BB, fold

Turn: (36.7 BB, 2 players) 7
UTG bets 18.3 BB, Hero calls 18.3 BB

River: (73.3 BB, 2 players) K
UTG bets 47.9 BB and is all-in, Hero calls 47.9 BB


Call light, make hands:

Yatahay Network - $0.10 NL FAST (6 max) - Holdem - 6 players
Hand converted by PokerTracker 4

BTN: 100 BB
SB: 251.7 BB
BB: 497 BB
Hero (UTG): 163.8 BB
MP: 290.3 BB
CO: 120 BB

SB posts SB 0.5 BB, BB posts BB 1 BB

Pre Flop: (pot: 1.5 BB) Hero has 6 6

Hero raises to 2 BB, MP calls 2 BB, fold, fold, fold, BB calls 1 BB

Flop: (6.5 BB, 3 players) 3 4 7
BB checks, Hero checks, MP bets 4.8 BB, fold, Hero calls 4.8 BB

Turn: (16.1 BB, 2 players) 7
Hero checks, MP bets 20.1 BB, Hero calls 20.1 BB

River: (56.3 BB, 2 players) 6
Hero checks, MP bets 50 BB, Hero raises to 136.9 BB and is all-in, fold



Raise preflop, make hands:

Yatahay Network - $0.10 NL FAST (6 max) - Holdem - 6 players
Hand converted by PokerTracker 4

BTN: 166.2 BB
SB: 110.4 BB
BB: 158 BB
UTG: 147.6 BB
Hero (MP): 120 BB
CO: 103.1 BB

SB posts SB 0.5 BB, BB posts BB 1 BB

Pre Flop: (pot: 1.5 BB) Hero has 6 6

fold, Hero raises to 2 BB, fold, BTN calls 2 BB, fold, fold

Flop: (5.5 BB, 2 players) 6 9 2
Hero checks, BTN bets 4.6 BB, Hero raises to 14.8 BB, BTN calls 10.2 BB

Turn: (35.1 BB, 2 players) A
Hero checks, BTN checks

River: (35.1 BB, 2 players) 3
Hero bets 24.5 BB, BTN calls 24.5 BB



Solver near 100% bets this hand on turn and when we check this turn it goes for the 300% pot overbet on the river with this combo. Would say I'm getting overly trappy but I think people are consequently overly tight in this spot. C/r, c/r is just too strong of a line though probably and not very balanced. Probably just need to bet big on turn, bet big on river.

3b light, get them to fold:

Yatahay Network - $0.10 NL FAST (6 max) - Holdem - 6 players
Hand converted by PokerTracker 4

BTN: 121.8 BB
SB: 249.7 BB
Hero (BB): 154.8 BB
UTG: 138.1 BB
MP: 156.3 BB
CO: 125.1 BB

SB posts SB 0.5 BB, Hero posts BB 1 BB

Pre Flop: (pot: 1.5 BB) Hero has J 9

fold, fold, fold, fold, SB raises to 2.5 BB, Hero raises to 7.7 BB, SB calls 5.2 BB

Flop: (15.4 BB, 2 players) T 6 Q
SB checks, Hero bets 4.9 BB, SB calls 4.9 BB

Turn: (25.2 BB, 2 players) 3
SB checks, Hero checks

River: (25.2 BB, 2 players) 4
SB checks, Hero bets 17.6 BB, fold


Also I forgot to mention I busted 5 hands in to the $2500 BOSS main after late regging level 3 with the ticket I won from The Beast. Lmao. AK < KK aipf 118bb bb vs HJ.

If you do well live but not online why not full ring online instead of 6max?

At this rate, I think trying to beat 50blitz is just a waste of time. There are many softer games available, and given ACR's track record I don't exactly have the highest level of confidence in their game integrity. While $50 isn't much money for most people, the eastern European regs grinding those games can afford food for a week with 1 buy-in. You can rest assured there are going to be players in that pool looking up preflop ranges and postflop solutions in-game.

I think playing online has given me an enormous edge in live games over the years.

Lately doing much better online, especially this year.

As for 6max vs full ring, on the internet it's mostly 6max games. The game starts from the Lojack position so basically take away 2 or 3 early position players. Conceptually it's not really any different, 6max allows you to play more hands over time since a UTG open shouldn't really be wider than 12-15% in a live 9 handed game, whereas LJ standard open should be at least 18%.

__

ddn - It's not a waste of time if it benefits my live play and helping me continue moving up in stakes there. If I can play against the toughest 50NL players, playing against a tough live pro at 10/25 or 25/50 doesn't scare me, especially if there are fish in the game. And since the latter has much more money involved I'd rather train on the former.