Latest cliffsnotes on Absolute soulreading.

[moving shapes]

yeah that's interesting. try and get a lot more players with vpip >50% over around 500 hands in your database, like someone said earlier. i think the only problem with these stats is that there's only one player with VPIP over 90% and very few above 60%.
i'm going to try and send you my database in the next week or so when i have time.

[Josem]

[ QUOTE ]
yeah that's interesting. try and get a lot more players with vpip >50% over around 500 hands in your database, like someone said earlier. i think the only problem with these stats is that there's only one player with VPIP over 90% and very few above 60%.
i'm going to try and send you my database in the next week or so when i have time.

[/ QUOTE ]

lol. i try and play with as many people with VPIP>50% as possible as it is.


More seriously, the problem is that people with very high VPIPs tend to go bust before they get to 500 hands - and thus, they get excluded from the analysis.

It is very unusual to find players with those sorts of stats simply because it is too expensive to play 500 hands in that style.

Of the 213 players with a VPIP of over 50% in this subset, they have an average losing rate of 31.7 bb/100.

To play 500 hands at that rate of losing costs $1585 at 5/10 NL, with no real prospect of winning.

[runout_mick]

Great work Josem.

I honestly expected the revised graph to have more deviaton, at least in a few of the more extreme sets. The closest is the point at 45%vpip 120bb/100, and that's still comparing apples to watermelons.

I think this is damning enough evidence to convince all but the most obtuse skeptics.

Case closed beyond any reasonable doubt IMO.

[Josem]

[ QUOTE ]
I honestly expected the revised graph to have more deviaton, at least in a few of the more extreme sets.

[/ QUOTE ]

Lol - the results cover around 350bb/100 divergence in win rates, and a difference from around 8 to 80 VPIP.

Keep in mind that the biggest "normal" winner is winning at 120bb/100 over 500 hands. I imagine that is huge deviation from their long term winning rate - as you would expect if you sampled over 2,500 people for 500 hands.

[ QUOTE ]
I think this is damning enough evidence to convince all but the most obtuse skeptics.

[/ QUOTE ]
I think the prior evidence is stronger in the sense of it being very clear.

This just provides that data graphically.

[ QUOTE ]
Case closed beyond any reasonable doubt IMO.

[/ QUOTE ]
Correct.

[apefish]

Josem- I don't think you need to revise any further.

The point of this was to graphically show how far out of the norm the cheating accounts were in relation to basically everything we know about online play.
The more you filter- the more you risk actually removing data that is relevant for comparison.

Many of the naysayers have said that surely somewhere at some time someone has run like this so therefore we cannot conclude there was something hinky going on.
In both instances the plot is so ridiculous that there can be no doubt about how out of line the results are.

Taking tags out of the equation for a future plot seems a bit of a stretch since the whole point is to show that these results cannot be replicated over X hands, not that one is much more likely to get certain results from playing a certain style.

Everything we want to see is in the plots you have done.
Creating it say for 400-700 hands with vpip over 50 only just shifts the center mass at the risk of not giving the overall picture a plot like this should give.


Solid job putting the data into an easy to visualize plot.

[flight2q]

How many hands did the blue dot in the lower left corner play? (Over 800 hands apparently, since not in the second plot.) What were the blue dot's postflop street aggression numbers? What was the standard deviation?

I gather you don't have a standard deviation for the red dot.

What happens when you filter for samples of around 500 hands at deep-stack tables only? Alternatively, what is the typical ratio of S.D. for deep stack and normal stack tables for players in your DB that play both? Which of the blue dots have an average number of players per hand that is similar to that of the red dot? Which of the blue dots are for stakes that are similar to red dots.

[Josem]

[ QUOTE ]
How many hands did the blue dot in the lower left corner play? (Over 800 hands apparently, since not in the second plot.)

[/ QUOTE ]
2,783 hands.

[ QUOTE ]
What were the blue dot's postflop street aggression numbers?

[/ QUOTE ]
Flop Agg 4.74
Turn Agg 2.32
River Agg 0.96

[ QUOTE ]
What was the standard deviation?

[/ QUOTE ]
I don't know how to calculate this.

[ QUOTE ]
What happens when you filter for samples of around 500 hands at deep-stack tables only?

[/ QUOTE ]
I can't. I don't have information on the stack sizes in the excel sheet.

The data that I do have is in the list (the inline image of text).

[flight2q]

Guess Holdem Manager doesn't report S.D. yet. That blue dot loses close to 100 buyins in 2783 hands. Not easy to do, especially with 22% VPIP. Maybe sees somewhat more flops than 22%, and makes big postflop bets. Hmm. Getting it in regularly with 25% equity postflop, 250 times and with 75% equity 50 times, results in S.D. of 280bb/100 from the all-ins alone. Possibly get something like that from shoving the flop with any pair or draw including two overcards, or something comparable that conforms to the street AF's. Probably S.D. is more around 400bb/100 due to variation in stack size and other stuff. That puts blue dot over 4 S.D. below breakeven (although hypothetical strategy is a big loser so actually just at expectation).

Red dot has S.D. of 520bb/100 calculated from a different 190-hand sample (deep stack) for one of the accounts. Possibly S.D. is different for the around 500 (hopefully unique) hands culled. Red dot typical preflop raise is to 3bb. Average first bet put into pot, when not reraising, is prolly around 2.5bb. Playing say 4-handed on average, the equity loss due to too loose preflop shouldn't be more than 1bb per hand (50PTBB/100). Actually, clearly less than this, because steals successfully quite a bit and play is positional from what I saw.

500 hands at some rate like 950bb/100, gives excess win rate of 1050bb/100. So that is between 4 and 5 S.D. to be explained by luck/skill/whatever.

Nice scatter plot. But usually number of deviations from expectation is what is plotted when different samples have different S.D.'s. Any real proof is probably not in the win rates over samples of this size. Any proof is probably in the unlikelihood of the postflop play being so successful; not based on the amount won, but on the details of the postflop play. As you said, for stronger evidence really need to go to the underlying data rather than such a plot.

[apefish]

It is however pretty solid for the "well it's really not all that uncommon or far-fetched" crowd- which is how it should be looked at.

[Josem]

[ QUOTE ]
Guess Holdem Manager doesn't report S.D. yet. That blue dot loses close to 100 buyins in 2783 hands. Not easy to do, especially with 22% VPIP.

[/ QUOTE ]

That "blue dot" lost $142.10. They seemed to play mostly 10NL at stars.

[ QUOTE ]
Red dot has S.D. of 520bb/100 calculated from a different 190-hand sample (deep stack) for one of the accounts.

[/ QUOTE ]
I am fairly confidence that is not what "standard deviation" means.

A standard deviation is how far results should fall within approx. 65% of the time when there is a standard distribution.

Your sentence seems to claim that 65% of the time, the Cheaters should have a bb/100 rate from -20bb/100 to 1000bb/100.

That is obviously absurd and meaningless, and I think it shows you don't know what a standard deviation is.

[ QUOTE ]
Red dot typical preflop raise is to 3bb.

[/ QUOTE ]
I don't know where this data claim is from.

[ QUOTE ]
Average first bet put into pot, when not reraising, is prolly around 2.5bb.

[/ QUOTE ]
I don't know where this data claim is from.

[ QUOTE ]
Playing say 4-handed on average,

[/ QUOTE ]
I don't know where this data claim is from.

[ QUOTE ]
the equity loss due to too loose preflop shouldn't be more than 1bb per hand (50PTBB/100).

[/ QUOTE ]
I don't know where this data claim is from.

[ QUOTE ]
Actually, clearly less than this, because steals successfully quite a bit and play is positional from what I saw.

[/ QUOTE ]
I don't know where this data claim is from.

[ QUOTE ]
Nice scatter plot. But usually number of deviations from expectation is what is plotted when different samples have different S.D.'s.

[/ QUOTE ]
wtf?

the sample is the bb/100 results and the VPIP results.

there are two samples here.

[ QUOTE ]
Any real proof is probably not in the win rates over samples of this size.

[/ QUOTE ]
as i noted above, winning at these rates is not concrete, 100% proof.

the charts show how obscenely outside normal parameters the accused cheaters are playing.