EVBB/100 vs BB/100

EVBB/100 vs BB/100

Do you guys think these metrics for win rate are interchangeable or do you think EVBB/100 is more accurate?

) 1 View 1
06 August 2024 at 07:27 PM
Reply...

48 Replies

5
w


Definitely BB/100 over EVBB/100.

Don't look at EVBB at all, would be my advice. Depending on your playstyle it can be very misleading/biased.


by plexiq k

Definitely BB/100 over EVBB/100.

Don't look at EVBB at all, would be my advice. Depending on your playstyle it can be very misleading/biased.

I tend to agree with this but I don't know the exact reasons yet.

In my PGC thread people are saying they prefer EVbb because it reduces a variance a bit or something along those lines.


Over large samples BB/100 is more accurate due to the bunching effect, card removal and collusion.


by ZKesic k

Over large samples BB/100 is more accurate due to the bunching effect, card removal and collusion.

Thanks ZK.

This does make sense to me. I intuitively thought BB/100 was a better metric but seeing the reasons laid out gives me more confidence in saying it.


Yeah, when you see those graphs with EVBB <-> BB winnings drifting apart over huge samples, that's much more likely to be caused by a bias in EVBB than being the once-in-a-billion hot/cold run that it looks like. I'd much rather look at a statistic that converges slower but reliably to the correct value.


I also prefer bb/100 over large samples.

Extreme example:

We are playing a game where we are flipping AdKd vs AsKs for 100bb. But my friend is colluding with me by removing 7d and 4d from the deck.
My win% is now 8%, while yours is only 4.4%.

My EV: 100*8% - 100*4.4% = 3.6bb = 360bb/100

Meanwhile HM/PT will say that evBB/100 is 0 and that I'm just getting insanely lucky, when it's actually just a matter of card removal.


wth yall yappin bout
yellow line is a closer estimation of your performance since it has lower variance than greenline, although not that much closer
on sites without rampant collusion that is, on those with tons of collusion you would expect to have a much better yellow line than green

You could maybe outperform your yellow line if you were extremely good at making "bunching effect" optimal decisions. But come on you don't make these decisions often enough for it to be reflected on your graph. Maybe I'm a noob but i can't remember one time i decided to go all in or not based on bunching effect. If you never account for it in your decision making, even if it's a real effect, it won't affect your graph because you will be off in a random way, not a directional one

by plexiq k

Yeah, when you see those graphs with EVBB <-> BB winnings drifting apart over huge samples, that's much more likely to be caused by a bias in EVBB than being the once-in-a-billion hot/cold run that it looks like. I'd much rather look at a statistic that converges slower but reliably to the correct value.

You should expect them to diverge bb wise, but to be proportionally closer as sample gets bigger in most cases. Not all cases though ofc


It's not necessarily about making conscious decisions dealing with bunching. Simply playing consistently looser or tighter than the population can easily be enough to bias EVBB - even without collusion.


by plexiq k

It's not necessarily about making conscious decisions dealing with bunching. Simply playing consistently looser or tighter than the population can easily be enough to bias EVBB - even without collusion.

Can you give a more specific example on this? It is interesting.


by aner0 k

wth yall yappin bout
yellow line is a closer estimation of your performance since it has lower variance than greenline, although not that much closer
on sites without rampant collusion that is, on those with tons of collusion you would expect to have a much better yellow line than green

You could maybe outperform your yellow line if you were extremely good at making "bunching effect" optimal decisions. But come on you don't make these decisions often enough for it to be reflected on your graph.

So you are choosing EVbb over BB/100?

The bolded doesn't happen that often but this is a clear example

Let's say we are in the SB with TT. We decide to RFI and BB3bets. We know UTG/HJ/CO/BTN folded which increases Ax/Kx in the BB's range (since it is less likely the folded players had one of those cards in their hand). We also unblock Ax/Kx with TT.

That means on average BB will be bluffing more when he 3bets. So now you always 4bet TT because 4betting will be higher EV even though a solver says its the same EV as calling.


by Zamadhi k

I also prefer bb/100 over large samples.

Extreme example:

We are playing a game where we are flipping AdKd vs AsKs for 100bb. But my friend is colluding with me by removing 7d and 4d from the deck.
My win% is now 8%, while yours is only 4.4%.

My EV: 100*8% - 100*4.4% = 3.6bb = 360bb/100

Meanwhile HM/PT will say that evBB/100 is 0 and that I'm just getting insanely lucky, when it's actually just a matter of card removal.

Cool example.


by plexiq k

It's not necessarily about making conscious decisions dealing with bunching. Simply playing consistently looser or tighter than the population can easily be enough to bias EVBB - even without collusion.

I think I get what you're saying

Villains ranges make the card distribution of the deck be weighted a bit more towards low cards, but then again bunching effect makes the deck more weighted towards high cards, so I don't know if the effect would end up being big enough to affect the graph.

I take it back, yall not yappin, very nice!

I still believe though when looking at samples that aren't millions of hands, yellow line will be better, but i can see how at some sample size green line would be the better one


by DooDooPoker k

So you are choosing EVbb over BB/100?

The bolded doesn't happen that often but this is a clear example

Let's say we are in the SB with TT. We decide to RFI and BB3bets. We know UTG/HJ/CO/BTN folded which increases Ax/Kx in the BB's range (since it is less likely the folded players had one of those cards in their hand). We also unblock Ax/Kx with TT.

That means on average BB will be bluffing more when he 3bets. So now you always 4bet TT because 4betting will be higher EV even though a solver says

That's not a good example, the solver knows the distribution of the folds, blockers and remaining cards in the deck on this node

I would say, even now after I understand the case for greenline>yellowline, it's only better at absolutely huge samples


by aner0 k

That's not a good example, the solver knows the distribution of the folds, blockers and remaining cards in the deck on this node

I would say, even now after I understand the case for greenline>yellowline, it's only better at absolutely huge samples

The solver is accounting for the 4 people who folded?

That is news to me, where do I find that metric on GTO Wizard?


by DooDooPoker k

The solver is accounting for the 4 people who folded?

That is news to me, where do I find that metric on GTO Wizard?

yes, it is. this is why if you look at a preflop solver for 3 handed, it will look different than the same positions on a 9 handed sim


by DooDooPoker k

Can you give a more specific example on this? It is interesting.

Suppose all players play exactly the same, except for one player who defends wider as BB vs SB. The BB range is already fairly wide there, so hands added will be predominantly low ranks.

All other things being equal - that player will now more frequently get into spots where they are all-in in a BvB spot and drawing to low ranked cards (or opponent drawing to high ranked cards). And since that's a spot starting out with 7 folds, the deck will be biased against low cards / towards high cards, so that player should end up running consistently "below EV" in terms of EVBB.


by aner0 k

yes, it is. this is why if you look at a preflop solver for 3 handed, it will look different than the same positions on a 9 handed sim

Oh you are just talking about a work around then. I'm talking about the solver actually showing you the bunching effect in real time regardless of how many people are at the table.


by plexiq k

Suppose all players play exactly the same, except for one player who defends wider as BB vs SB. The BB range is already fairly wide there, so hands added will be predominantly low ranks.

All other things being equal - that player will now more frequently get into spots where they are all-in in a BvB spot and drawing to low ranked cards (or opponent drawing to high ranked cards). And since that's a spot starting out with 7 folds, the deck will be biased against low cards / towards high cards, so t

Yeah that makes sense to me.

So let's compare heads up to 6max. Would you use EVbb for heads up and BB/100 for 6max since there is no such thing as bunching or collusion in a heads up game?


For HU EVBB is clearly superior, yeah.


by plexiq k

For HU EVBB is clearly superior, yeah.

Sweet okay I understand it now, thanks plexiq.


by ZKesic k

Over large samples BB/100 is more accurate due to the bunching effect, card removal and collusion.

Can you provide a realistic example of a hand where collusion would reduce the number of outs, thereby lowering my EV? Additionally, when comparing EV to BB/100, we typically discuss larger discrepancies, such as +/- 2BB or more. Do you believe that bunching effect, card removal, and collusion have a more significant impact on EV loss than these larger swings?


by plexiq k

Suppose all players play exactly the same, except for one player who defends wider as BB vs SB. The BB range is already fairly wide there, so hands added will be predominantly low ranks.

All other things being equal - that player will now more frequently get into spots where they are all-in in a BvB spot and drawing to low ranked cards (or opponent drawing to high ranked cards). And since that's a spot starting out with 7 folds, the deck will be biased against low cards / towards high cards, so t

Say there's fish defending 80% of hands in the bb vs sb open - how much will this actually change bb vs evbb?

Intuitively it feels unlikely to change it much given the frequency of getting all in BVB SRP plus the actual change in odds from card removal but could easily be convinced otherwise.


by ZKesic k

Over large samples BB/100 is more accurate due to the bunching effect, card removal and collusion.

Bunching effect and card removal should even out over time no? Maybe you're a bunching effect god and that gives you an extra .05bb winrate not shown


by Meeowth k

Say there's fish defending 80% of hands in the bb vs sb open - how much will this actually change bb vs evbb?

Intuitively it feels unlikely to change it much given the frequency of getting all in BVB SRP plus the actual change in odds from card removal but could easily be convinced otherwise.

No idea about the absolute impact, I'll run some sims. Shortstacked or ante games probably make for a better example here, but going with regular 100bb.

Reply...