Standard 3barrel Spots

Originally posted by w34z3l
Hey guys,

Now we have some empirical evidence to show that 3-barreling as a bluff can be a nice way to boost winrate and red-line even at lower limits.

I can't get my head around why this graph supports 3barrel bluff profitability. Showdown hands are slightly higher than non showdown hands. Wouldn't this seem to indicate that slight -EV given we never have a hand at showdown?

It's a fair point, and essentially increasing our red-line through 3barreling will have a negative effect on the blue-line.

However, notice that there are more flat sections on the blue-line than the red-line. (Red line appears to be more of a fluid curve where the blue-line has more "corners".) This generally means that most of these hands didn't reach showdown. In other words our fold-equity should outweigh the cards we are holding at any point.

The only effect our cards will have on a non-showdown situation is blocker effect etc. Not to mention if we 3barrel bluff every time then some players may decide to call us down wider. But essentially it seems that 3barreling will increase our red line more than it makes our blue line decrease, resulting in overall positive effect on green-line. If our initial assumptions are correct we would not expect to see the red-line positive at all.

Notice also that this graph does not automatically convey that 3barreling any 2 cards is profitable in the long run. Naturally this graph does not show any situation where we attempt to 3barrel and perhaps get raised on the flop and turn. But it does show us that assuming we reach the river with air, firing will be more profitable than checking in the long term. This is actually also in harmony with GTO principles to an extent. It's usually correct for the caller to slightly over-fold the river since he has an inherent range disadvantage. (It's common to see fold to river cbet stat of significantly above 40%, which is not incorrect from a GTO point of view). It's one of those situations that we should always make auto-profit vs a decent opponent. He defends against this not by calling wider but by making sure he 3barrels with an appropriate frequency in a similar situation where we give him auto-profit. So if we have strong 3-barrel fundamentals and our opponent doesn't, we should show long-run profit.

I follow your concepts, but I don't see "empirical evidence"

This might be a dumb question, but as I don't use tracking software, could you explain the meaning of those different lines?

Green = Profit
Red = uncontested pots
Blue = showdown pots

Originally posted by GoOnCal1
I follow your concepts, but I don't see "empirical evidence"

Feel free to dig deeper and use a graph with larger sample.

You could also use notecaddy or player-aliases to calculate average fold-to-river-cbet stat. If it's noticeably above 40 then that's more evidence to suggest auto-profit with 2/3rds pot size bet on river.

Hi Adam,

I think that looking at red/blue line for determining how successful is 3 barrel bluff can be missleading. As I understood, you filtered hands where hero did 3 barrel, and the graph came out as a result? Can we expect in that case, red line to be negative? Lets say villians calls too often, 80% vs 3 barrel in average. When they call, it will go to blue line, when they fold it will increase our red line. The only way that hero's red line will decrease, is when hero bets and villain raises and hero folds. There will be also cases where hero bets, villian raises, hero reraises, and villian folds which increases also hero's red line. Since raises on river are in general very rare, I think it is obvious that red line will be very positive.
In your case, blue and red line are somehow similar. What does it mean? It can mean so many things (e.g. villians mostly call and hero is bluffing too often so blue line is not great, and since villians rarely fold red line is similar as blue). My point is, looking at blue and red line for 3 barrel spots (and most other spots) is not helping us to conclude how successful is our line.

Imo, more accurare approach would be to make a filter for 3 barrel, and set hand category to high card or weak pair (which are almost allways bluffs in 3 barrel spots, could also add mid pairs, but it doesnt matter much). Then we can look at stat "cbet river success", which shows how often is our 3 barrel bluff successful.
Then if we can look the same for value hands by negating filter for "bluff hand category". This way, you can see for any line, how successful is bluff and how good is value hand. If you want to go in more details, you can add filter for board category type, and e.g. see how often is bluff good on draw heavy board which is the case for this board.

Regarding the JQo hand, theoretically not a good bluff candidate since it blocks busted draws and has showdown, also it blocks KJ, KQ, but it is the question if it is in limped range, and it also blocks TJ, QT, JJ, QQ which will villian fold.
It totally depends what is % of busted draws in villian's range, and if he folds Tx, JJ, QQ or even Kx. So I would make decision here purely on river cbet success vs population, (or vs villian if enough sample), and from my experience, population on micro limits (even NL4) is overfolding flop, overcalling turn and folding around 40-45% river, so I wouldnt bluff this hand especially because it has showdown value (if you thin value bet Kx and cc river on such board, villians are perhaps not going to bluff you that often, and if anyway they bet most of their bluffs, you can start cr bluff since they really dont rep much besides busted draws when they bet river). I would bluff only lowest busted fd, since they block least fold range in villian range. I would decrease my bet size to 50% on river, to make a bluff +EV. if villians are not bluff raising and induced, then to 33%.

This?
EDIT: doesn't apply to limped pots tho - cbet stat is only when you raise PF I'm guessing.

Need to add made pair one pair is bottom pair I guess too - I bluff blockers a lot, turn bottom of range into a bluff later

What does this say tho
because if I check
The graph will show what I lost last 2 streets as well so can't compare to EV=0 and can't do what I do e.g. in BB when looking at -EV calls as -100bb/100 being 0EV

So I guess put this in just doing this and then going to cbet success

EDIT: 49.55%
So I am the master grandmaster river bettor champ - average sizing will be like 65%
So yeah I should look to barrel my air balls more often thanks w34z3l and GingerKid!

Okay added more stuffs

with blockers and under pairs (combined with previous)
it comes up to 51.34% R cbet success 261 hands

Should really bluff more

Blocker lower pairs work more often than high cards (not surprising + small sample tho)

1.1 million hands database btw guys
Keep these coming I will give my input
Since My PC is slower than Dwan playing live I might need to do a purge soon

As to this hand, whatever, people will overcall bullshit draws especially limpers so we can bet regardless of blocking QJ imo - this is a limped pot at a limit where people overcall FDs especially to small dollar amount, so low flush blockers do play indeed vs limping/completing ranges (not getting a boner over the Qh tho :p) . Like Ginger Kid's half pot or less to get the bad boys fold those and just let them call their Kx Tx - Sometimes you get called by Tx anyways by limpers at NL5 when you bet like this, happens quite a bit actually.

Originally posted by w34z3l
It's usually correct for the caller to slightly over-fold the river since he has an inherent range disadvantage. (It's common to see fold to river cbet stat of significantly above 40%, which is not incorrect from a GTO point of view). It's one of those situations that we should always make auto-profit vs a decent opponent. He defends against this not by calling wider but by making sure he 3barrels with an appropriate frequency in a similar situation where we give him auto-profit. So if we have strong 3-barrel fundamentals and our opponent doesn't, we should show long-run profit.

Great quote
The 1-A poker simplicity was fun while it lasted

Thinking more about bluff spots, has made me smile more this week.
Thinking about developing empirical data has had me distracted.

Bluff experiment
Bot profile A does not bluff with no hand on the river
Bot profile B does, else both players range and decision trees are identical.
Both players play a significant number of hands in the same arena
Compare results

Originally posted by GoOnCal1
Thinking more about bluff spots, has made me smile more this week.
Thinking about developing empirical data has had me distracted.

Bluff experiment
Bot profile A does not bluff with no hand on the river
Bot profile B does, else both players range and decision trees are identical.
Both players play a significant number of hands in the same arena
Compare results

Wouldnt regs overfold vs bot A since they would notice he is value heavy ?

Originally posted by LemOn36
Okay added more stuffs

with blockers and under pairs (combined with previous)
it comes up to 51.34% R cbet success 261 hands

Should really bluff more

Blocker lower pairs work more often than high cards (not surprising + small sample tho)

1.1 million hands database btw guys
Keep these coming I will give my input
Since My PC is slower than Dwan playing live I might need to do a purge soon

Your filter is good, I am using the same. And yes, you should bluff more keep in mind, it can be that regs overfold vs you because they noticed your low river cbet frequency, or your low bluff river frequency, which can change if you suddenly start bluffing. If you want to use a version of pt4 bluffing frequency, you can download custom stat "river bluff frequency", which is a great stat to have, can put it in your HUD.

Link pls to the stat / custom stat thread?

https://www.pokertracker.com/custom/stats

There are a lot of nice custom stats. It will slow down your hud since custom stats are not well optimized as default stats, but it is ok. I never tried it with huge databases.

By the way, since you have big data sample, did you filter bluff river success, on dry board, wet board, paired board, A high, K high, low board? It would be also very interesting, because some spots are very bad for 3 barrel, and some are not. Depends a lot on population, e.g. on Axx many regs will fold very little vs 3 barrel on lower limits, and on e.g. Nl50, Nl100 it could be that they overfold.

Originally posted by GingerKid

Originally posted by GoOnCal1
Thinking more about bluff spots, has made me smile more this week.
Thinking about developing empirical data has had me distracted.

Bluff experiment
Bot profile A does not bluff with no hand on the river
Bot profile B does, else both players range and decision trees are identical.
Both players play a significant number of hands in the same arena
Compare results

Wouldnt regs overfold vs bot A since they would notice he is value heavy ?

Sorry I'm just being pedantic about the term "empirical evidence" and suggesting that clinical trials of a proposed hypothesis is what this term means,
as opposed to the rationalistic perspective (or argument) backed up by data mining, which aint "proof".

Originally posted by GoOnCal1

Originally posted by GingerKid

Originally posted by GoOnCal1
Thinking more about bluff spots, has made me smile more this week.
Thinking about developing empirical data has had me distracted.

Bluff experiment
Bot profile A does not bluff with no hand on the river
Bot profile B does, else both players range and decision trees are identical.
Both players play a significant number of hands in the same arena
Compare results

Wouldnt regs overfold vs bot A since they would notice he is value heavy ?

Sorry I'm just being pedantic about the term "empirical evidence" and suggesting that clinical trials of a proposed hypothesis is what this term means,
as opposed to the rationalistic perspective (or argument) backed up by data mining, which aint "proof".

Yes, such experiment would be the best estimation of how good is bluff 3 barrel, but still it is not a proof, unless they would somehow play identical hands, and villains play identical vs both bots. Since in reality it is not possible vs real players, then such experiment is not going to show us much, since real players are going to play differently vs bot who never bluffs compared to the bot who does bluff. Also data mining can be misleading, but imo it is the best we have in the real world

I quite disagree

Paired trials, in real environment's, provides

Originally posted by GingerKid

Originally posted by GoOnCal1

Originally posted by GingerKid

Originally posted by GoOnCal1
Thinking more about bluff spots, has made me smile more this week.
Thinking about developing empirical data has had me distracted.

Bluff experiment
Bot profile A does not bluff with no hand on the river
Bot profile B does, else both players range and decision trees are identical.
Both players play a significant number of hands in the same arena
Compare results

Wouldnt regs overfold vs bot A since they would notice he is value heavy ?

Sorry I'm just being pedantic about the term "empirical evidence" and suggesting that clinical trials of a proposed hypothesis is what this term means,
as opposed to the rationalistic perspective (or argument) backed up by data mining, which aint "proof".

Yes, such experiment would be the best estimation of how good is bluff 3 barrel, but still it is not a proof, unless they would somehow play identical hands, and villains play identical vs both bots. Since in reality it is not possible vs real players, then such experiment is not going to show us much, since real players are going to play differently vs bot who never bluffs compared to the bot who does bluff. Also data mining can be misleading, but imo it is the best we have in the real world

I quite disagree
Paired T tests in a real environment provides a solid empirical evidence.
You could quite easily run a model with identical hands as well
Perhaps on snowie or other similar tool


Later adds...
Don't get me wrong, I don't disagree with the OP's premise.
I have been bluffing all week, so I can provide observational and anecdotal evidence
In the real world lots of people are building better bots and the best way to improve them is testing in the real world.