If baccarat shouldn't be a so complex model, the most ambitious goal would be to approximate at best the actual patterns distributions of EVERY shoe dealt.

I'm not joking, as long as the quality of streaks (and singles) is somewhat limited, the first/intermediate portions of the shoe should give us slight hints capable to erase and invert the HE in our favor.
And of course one of the most important thing to assess is the shuffling procedure employed at that particular casino you're playing at.

To do that, we need "complex" patterns to take as possible triggers because more hands are needed to form a pattern (let alone two consecutive patterns...etc) lesser will be the whimsical third card(s) impact over the results.

Since each shoe features an asymmetrical card distribution by any means, on average we'll expect more clustered patterns of any kind than "alternating" patterns, yet alternating patterns must show up sooner or later because they have to somewhat catch the most probable clustering propensity.
Thus betting is a delicate process directed to get the most at pattern clusters and the least damage at alternating patterns.
In reality even the "alternating" model is expected to provide unlikely "overalternating" sequences that could be bettable as well.
 
Curiously, people invited to write down "random" sequences at binomial games (e.g. coin flips) are more prone to provide (wrong) overalternating sequences.
And most part of baccarat players tend to reverse such propensity thought, thinking that the alternating mood is too much subordinate to several kind of clustering patterns.

More later

as.

That's what I'm trying to say.

Google "wheather randomness" and the first words are those:

If a physical system is sufficiently complex then it also will exhibit randomness.

Well, baccarat is everything other than a complex model as outcomes are the by product of innumerable card combinations but merging toward more likely patterns formation.

That is running several times the same situation, results will be surely shifted towards one side by different levels that are in partial relationship about how things went so far in the actual shoe.

It's the "partial" relationship factor we should focus upon but not forgetting that itlr some events are more probable than others.

I've already sayed that by setting up a "approximate algorithms" plan, the probability to lose is virtually zero.
And just for the reasons you've illustrated in your post above.

Differently than most part of human minds, algos realize quickly how many "profitable" patterns should come out per each shoe dealt, especially after many hands are dealt.
Of course the word "profitable" assumes a relative weight, but whenever things seem to be unclear the best option to make, by far, is not to bet.

Anything can happen no matter what?
Sure, but by limited standards.

Say we want to pick up randomly three numbers from 1 to 28, then checking the pattern corresponding to the column number of any shoe we have collected.
To simplify the problem, say our "enemy" will be a 5/5+ long streak, singles are considered as neutral, so "enemies" must fight vs 2, 3 and 4 streaks.

For example numbers picked up are #3, #11 and #22.
What's the pattern happening at those column numbers?

Odds are that at most circumstances no 5/5+ streak happened, even if it's way likely such columns were filled by neutral singles.

Whenever two or three streaks of any nature really happened at those randomly picked up spots, we should consider if any kind of "more probable" clustering effect should be working, first by considering 2/3 long streaks, then 3/4 long streaks and finally by a 2/4 long streaks distribution.
A less interesting scenario is to see which streak followed a possible 5/5+ streak happening in the first position or in the second position if the 3 picked up numbers corresponded to three streaks.

Repeat the process infinitely and you'll see that it's way more probable to get 2/3 and 3/4 streak clustered than any other scenario, even knowing that many times you are not in the position to bet anything (for streaks not happening at least twice within the 3 columns range).

I understand that this is a sort of intricate example, but it should get the idea that bac results are not perfect randomly distributed even we had implemented at the start a random factor.

as. 

Thanks KFB for your post.

Please can you elaborate your interesting Random doesn't  mean scattered evenly passage?
Thank you!

Anyway I think that a succession should be taken as really random (so not depicting any "valuable" pattern along the way, that is a unbeatable sequence) when the limiting values of relative frequency remain costant no matter the points of the succession considered.

Obviously each possible pattern ITLR will get the proportional values expected by math, yet we should be more interested about WHEN and HOW MUCH different points of the sequence will make room to some detectable patterns.
In this way baccarat is definitely a unrandom game.

This way of considering bac outcomes get rid of most part of "averages", focusing more about actual patterns probability of coming out "out of blue" or actual patterns more likely lenght.

To get a better idea of that, suppose a casino voluntarily or coincidentally provide two or more consecutive shoes not belonging to averages, so taking a unidirectional unexpected line for long.
In relationship of the actual method we're using, such situation could be either a heaven or a hell and to make a living at this game we ought to know that heaven and hell are two extremes we should try to avoid at all costs.

More later

as.

Hi Al!

Here is:

Without any doubt and by playing a lot of hands, the number of profitable/unprofitable situations stays way below of the 1/1 ratio and actually it should remain unfavourable even when we bet five, two or one hand per shoe or just one hand per every ten shoes dealt.

I'll try to elaborate such concept.

Math teaches us that any bet is EV- no matter what, so no human influence (or fkng mechanical models or progressions) can invert it and this is a utterly indisputable statement.
Whenever you bet 1 to get a 0.9894 or 0.98.76 return, you're losing money itlr, period.

Of course math assumes that bettable successions are randomly and indipendently produced.

Baccarat literature has never investigated whether bac successions are really random, neither about how the "dependency" factor could be measured as both parameters were simply ascertained as 1) a sure feature (all shoes are randomly produced) and 2) any new hand is completely disjointed from previous hands.

Since black jack was found to be a beatable game (well before E. Thorp published his book), "experts" tried to apply the same math features at baccarat, obviously with no avail.

Whereas bj successions can get a lesser fk about a possible unrandomness, so focused about the current high cards/low cards ratio, at baccarat ALL successions were and are considered randomly produced.
Actually a kind of baccarat dependency was spotted, but acting by insignificant values.

So under the eyes of gambling experts baccarat remains a random EV- game.

This is a 1 billion false statement, such people didn't know the best definition of randomness ever made, let alone how much a finite slight dependent model will act by transforming it into a unrandom sequence.

Casinos do not know a fkng nothing about this, they just collect the profits as people keep playing the game without really knowing what to look for.

Therefore people willing to open the door about the bac vulnerability are considered just as clowns, unless they'd bet huge sums and being consistently more right than wrong.
Now math laws as well as mathematician assumptions start to be debatable, to say the least.

Let the house getting its math edge, we'd get the best of it no matter what.

as.

Thanks for your replies Al and KFB!

Of course when an opportunity comes around, well we should try to exploit it, even if we have chosen to set up our plan in order to win  every single session we play (that is an average profit per a given number of shoes played). That means to let it go (without wagering) some long profitable situations.

Without any doubt and by playing a lot of hands, the number of profitable/unprofitable situations stays way below of the 1/1 ratio and actually it should remain unfavourable even when we bet five, two or one hand per shoe or just one hand per every ten shoes dealt.

Whereas the former part of the comment above relies upon common sense and experience (and math), the second part rely upon math, that is by assuming a total randomness and independence of the outcomes being always EV-, a thing completely disappointed by our studies.

In fact there's no one shoe composition in the world getting multiple random walks applied to the same succession getting "more likely" steps be silent for long or not forming sequences of a certain lenght.

As long as a shoe is formed by finite decks and cards are burnt after each hand resolved, things must take one direction or another by measurable values.

See you later

as.

Yep, I was talking about unfavourable opposite events, I'd guess your shoe is a paramount example of a strong FAVOURABLE situation to get the best about. 

BTW, you can't imagine how many posts of yours have improved our betting model.

as. 

The primary tool why this game could be beatable itlr is because any shoe distribution is "biased" at the start, that is a strong asymmetrical force will always and constantly act as a decisive factor between the math issues and the actual (asymmetrical) distribution.

Of course per each shoe dealt such force will work by different quantities and (more importantly) qualities that for practical reasons we had to condense in the pattern formation and distribution.

Simplifying, what didn't happen so far in the actual shoe is considered as "not existent" by our algos, providing a proper room of apparition was left in relationship of the number of hands dealt.

An event not happening so far cannot be classified as "isolated" or "clustered", the secondary main tool why we should win at this game.
Best example to provide is any streak not belonging to a specific class and we already know that we can safely assume that streaks could be restricted within 2, 3, 4 or 5/5+ classes.

Therefore any FOUR streaks sequence happening (streak cutoff= two hand in a row) is more likely to form a kind of clustered succession among 2/3, 3/4, 2/4 classes than any 2/5, 3/5 or 4/5 event.
Obviously it'll happen that ALL consecutive streaks are belonging to the 5/5+ category or that any 2, 3 or 4 streak will be intertwined by a 5/5+ streak.

In some way and even if one doesn't know the exact pace our algos work on, probability to get all different streaks after four streaks happened is quite low.
No matter what and since we do not know the exact "asymmetrical" factor strenght actually working, when 4 different numbers are coming around, it's 100% certain that at least one streak 2,3 or 4 cluster must happen.

Even when two unlikely 5/5+ streaks show up within a four streaks range, some inferior streak classes must come out clustered. Most of the times belonging to the 2/3 streaks category, then 2/4 category and finally to the 3/4 category.

In reality, such streak clustering effect might be diminished (or even erased) when a low streak/single ratio happens in relationship of the number of hands dealt so far, most of the times when long chopping lines come out consecutively.

After all, every pattern considered by a number will fight vs an equal or superior/lower number as we are taking into account a 0.75% general probability to succeed.

It's completely obvious that longer streaks will come out more isolated than clustered, and when they are not most of the times is because a shortage of streaks happened so far.

On the other end, when many streaks dominate the model, it's almost impossible to miss a clustering effect of some kind.

We have even set up a progressive mechanical betting after waiting that any 2 or any 3 streak will come out as isolated three or four times in a row.
Very rare situations to happen, it's like to wait a strong and favourable positive bj count.

The difference is that at baccarat we are not forced to bet a fkng dime, just let the house to confide about improbable events to happen for "long".

as.

"Bad" shoes predominance

Suppose the casino knows perfectly what we're doing, that is exploiting the "average" shoe texture thus starting to offer "strong opposite deviated" shoes for long.

Differently to what many could think of, it's quite easy to arrange the cards in order to form a kind of "distant than average" shoe composition, shoes where the cut or the initial cards burnt in relationship of the value of the very first card cannot alter anything other than few patterns.

It's just an hypothetical consideration, almost every live casino in the world acts by perfect legal standards.
But for example, think about a machine malfunction (shoes not being entirely shuffled) or other manually shuffling issues.

We'll expect "average" but we keep facing an unfavourable "deviated world".

Now we need to find the best compliance to the shoe(s) we're facing, meaning that one or a couple of favourable triggers coming out after a world of disaster do not necessarily mean to start the betting. And of course at most circumstances trying to get the best of it by exploiting a unlikely negative deviation to prolong represents a thing our algos are not interested about.

It's true that at the most part of shoes algos will spot a slight greater amount of positive situations than negative spots, yet around a 2%-6% percentage of total shoes dealt will pose a real threat for us, so wiping out most or all of the previous profits we had accumulated.
So even a careful detailed clusters and isolated pattern evaluation could be of no avail for our plan.

Technically those shoes are just a natural occurence destroying any mechanical method ever invented and, on the other end, not easily controllable by a simple human adaptation.

More later

as.

The main tool algos are interested upon is not the probability to win a lot of times or to hit several "freerolling situations" (that is patterns reaching or best surpassing the 3:1 ratio), but to restrict at most the natural negative variance where the "model" seems to be too unpredictable (or deviating too much from the "norm" at the unprofitable side).

Check your shoes data whatever you want, eventually you'll see that when a proper random walk is working the vast majority of shoes dealt will produce from one to three 5/5+ streaks, so you should find betting ways to get the best of it at those more probable circumstances.

Then when things seems to be too deviated at "unexpected" side of the operations, algos will raise  the betting parameters by waiting, for example, two or three fictional wins up to the point where they are not interested to make any bet for that shoe.

In summary, W long clusters should be "gambled" just in very few situations and anyway always by not jeopardizing a previous profit or in order to recover promptly an actual loss, thus this is the exact opposite way recreational players like to do.

Remember that assigning a 0.75 W probability, W clusters are real winning clusters only when they surpass the 3:1 W/L cutoff point, or 2:1 or 1:1 or worse W/L ratios should be considered as natural losing spots we can't do anything about. (Actually we've seen that any W cluster--as WWL sequence---might be good but not "so good").

So consider any  WWL sequence as a kind of "backup" plan.

W long clustered sequences vs L long clustered sequences and W isolated clustered events

We've seen that some random walks working by a 0.75% probability of success itlr will form longer W successions than a proportional amount of L successions. (W=+1 and L=-3).
We've found out that the gap, albeit being rare to cross through, will provide us a strong statistical advantage.

Cutting to the chase the issue, rare situations on both sides of the operation will proportionally privilege our 0.75% probability and not the 0.25% counterpart.
Such propensity slightly decreases in relationship of the W pattern lenght up to the point where even W isolated events will be more restricted in their back-to-back apparition than what math dictates.

In fact, after any couple of isolated W events the most probable outcome will be WW and not WL by percentages well surpassing the general probability (0.75%), getting winning percentages ranging from 78% to 80%.
That means that in those (relatively rare) situations our edge (before vig) will get values up to  20%.

This is just one example of the #2 point discussed above.

See you next week

as.

That's the beauty of baccarat!

Two of the best bac scholars (Al and KFB) have chosen two different options (1 and 3 or 2).

Let's see what we had implemented in our algos to "solve" the puzzle.

Live players action

First, most of the live action worldwide is made by hoping that something will prolong several times in a row, think about tourists and the vast majority of gamblers.
Nothing wrong with it, providing to maintain low or very low the betting pace as long as W "homogeneous" clusters are correctly assessed by a general statistical point of view.

So for example, it's quite difficult to face a shoe NOT forming singles/doubles sequences for less than 4 spots. But at the same time it's quite difficult to cross shoes not forming a 2 or 3 s/d sequence. (Not mentioning those isolated s/d spots). 
Since any 3/3+ pattern will make us a -3 unit loser, after a win or a couple of wins we should be worried to concede the previous profits to the house.

Technically we need at least a 4 s/d sequence to gain a profit and of course not every succession will accomplish this.

On the other end, odds that a s/d sequence will reach the 6 or 7 or superior value are quite good.
Yet, it's not so probable to get such sequences more than one time per any shoe dealt.

Simplyfing, the W clustering effect of decent lenght will be somewhat limited to one time per shoe.
Or, it's the same concept, that s/d sequences surpassing the 3:1 cutoff value normally are not overwhelming the 3:1 overall profitability, meaning that most s/d successions won't provide back-to-back (consecutive) numbers greater than 3.

More later

as.

Quote from: KungFuBac on June 03, 2024, 02:40:56 AMGood post AsymBacGuy

Personally I find #2 easier to discern vs #1 or #3:

"2- Considering it to stop a L pattern of given lenght"


Continued Success,

  Thanks!
Of course I knew you "chose" the most reliable point to look for.

as.

Summary

A(p)= 0.75% and B(p)= 0.25%, where (p)=math probability

A utopian world would produce successions as AAABAAABAAABAAAB....

Actually the vast majority of bac successions won't provide such distributions for long other than by a kind of unlikely strong "coincidence probability", so we'll expect that the vast part of shoes dealt will diverge (in a way or another) from such "utopian" pace.

Notice that differently than other perfect random independent successions (e.g. EC roulette outcomes) such  world will be somewhat "biased" at the start for the sure undeniable asymmetrical card distribution and for the bac rules favoring B side.
Of course we do not know which A or B side of the events will be favored at the various portions of the shoes and by how much.

Suppose A= searched (W) spots and B= unfavourable (L) spots

A succession as AABAABAABAAB...would be altogether beatable despite of performing a strong shifted transitory probability privileging B side. In fact now A=0.666% (instead of 0.75%) and B=0.333% (instead of 0.25%).
Who cares? AA still remains the best option to make a singled A bet.

At the same time at such two different scenarios, B events remain isolated so it's a child joke to  get them coming out as isolated and not clustered.
Unfortunately and by those precise ratios (3:1) the first "utopian" succession won't happen for long, yet the second one (2:1) is way more likely to succeed as it'll be mathematically more likely to get any kind of A cluster than an A cluster surpassing the AA cutoff.

Obviously any A cluster surpassing the AA cutoff point will get us a win and for the reasons already traced, we're entitled to get some superior AAA patterns than precise AAAB patterns.

But who knows?
It's better to secure a win after any A(A) situations than hoping to get a kind of sky's the limit AAA...sequence where a single loss will wipe out three wins.

In a word, a s.t.upid plan oriented to get A clusters of any kind will suffer the least impact of negative variance.

On the other side, B events should come out more isolated than clustered but someway they must catch up (balance) the possible more likely math propensity to get LONG A clusters (a thing will see in the next post). Thus coming out more clustered than isolated.

Again, a utopian world would be to face long successions of B isolated spots, then two B clustered  spots.
But since the model is strongly asymmetrical, we can't rely upon precise values so we might add the factor of any A situations intertwined by any single B event. So we are not interested to bet toward A when B keep showing up.

In practice and considering a given random walk or multiple common random walks, our large live shoes sample had shown us that A probability to come out clustered doesn't remain constant after two A events coming out as isolated. That is after a couple of A isolated spots, AA will overwhelm the 0.75/0.25 probability ratio so getting profitable values well greater than 0.75.

Obviously some could argue why a BBBBA...succession won't get valuable A bettable spots than a B..AB...AB...A...sequence where now A is way more likely than B.

The answer is that the greater two initial cards point is 2:1 math favored to win the final hand, but it's sufficient to get one hand going wrong to alter the more likely A/B pace and when results keep staying to one side of the operations, we'd better wait for two "fictional" A losses not displaying a more likely course of action.

I've already sayed that (no matter how's the random walk utilized) long streaks are the mixed product of 1) unlikely "long" consecutive greater two initial cards points and 2) math two initial card underdog points getting a favourable third(s) card impact.

Basically and at least after having studied our large live shoes sample, we've found out that the more likely two initial greater point will get a two value pace, so we dared to reach the conclusion that at baccarat doubles are the more likely results for this reason.

Of course a large part of outcomes will disrupt such allegedly propensity, that's why we had to implement a so called "multiple variables" factor in our plan.

as. 

Winning spots

There are three different ways to catch a W spot:

1- Hoping it'll come out as a "starting" spot

2- Considering it to stop a L pattern of given lenght

3- Considering it to come out again after a W spot (W cluster)

From a math point of view, such list doesn't make any sense as the W probability will remain 0.75 indipendently of what we're trying to dissect.

Actually (and fortunately) things doesn't correspond to those raw probabilities for each W scenario happening.
In fact even the first starting scenario mainly based upon "luck", itlr will form detectable W distributions needing quite time to be correctly grasped, many times by letting go those natural L clustered events to show up.

The third point seems to be quite straightforward but it is not, we reckon being the #1 reason why most players fail.

The second point needs a lot of time to provide real bettable spots, but by far will provide those sure strong EV+ situations completely denied by math (but not by statistics).
Remember that we do not want to win 100% of those relative rare allegedly EV+ spots, we'll always expect negative variance putting us into a harsh emotional status.

Yet, whenever a verified situation launched "infinite" times won't provide proportional math values, well we'd think to be in a very good shape to exploit an advantage.

More later

as.

Winning patterns and losing patterns

Think deeply about this: If you want to be ahead of something (other than by short term luck) you need to get a pattern winning at least two times in a row or to come out at a selected spot after a given sequence of losses (being one, two, three or more).

Technically that means to spot a kind of homogeneous (albeit short...that is just two events) step movement towards the winning side or to stop a given losing sequence by one step back, so an alternating W/L succession could come out at our favor too, even if the asymmetrical 0.75/0.25 probability will cause more units lost at L sequences than winning units at W sequences.

The natural W sequences being more probable than two steps just add more fuel at our plan. But in order to avoid the fkng negative variance, it's better to play our main bets towards W sequences being clustered just one time or, at a lesser extent, to limit L sequences to show up as isolated (singled) and not hoping to get long W spots around any corner.

Nonetheless we'll see that waiting patiently (meaning to assess some shoes dealt) some specific losing lenght successions to show up, our edge over the house will be astronomical.

Basically at the first degree of operations our algos work about those simple situations: (cl=clustered and isol=isolated)

W(cl) > W(isol) (main tool)

L(isol) > L(cl) (secondary tool)

So a valid random walk to be reliable must get:

W(cl) + L(isol) > W(isol) + L(cl)

There's no way that itlr such proportion will be disregarded for long as a 0.75 probability overwhelms the 0.25 counterpart, even if we use the most vulgar bet selection.

Yet, some quite rare consecutive shoes will pose a threat about that, forcing us to choose which of the two elements (Wcl or Iisol) will get the best probability to confirm the above main assumption.

That's the secondary level of algos "thought".

Given for granted that (albeit being relative rare) consecutive W(isol) plus L(cl) sequences are very harmful, algos are somewhat "forced" to choose which element will be more silent at the remaining portion of the shoe and of course they set up a higher level of both W clustering effect and L isolated effect.

Thus now the W clustering effect comes out as effective just after TWO fictional winning spots, the L sequences by now remain "undepictable" as long as they won't concede room to the more likely W clustered sequences.

Therefore given a 0.75/0.25 W/L probability, whenever a relative unlikely WLL sequence happens, algos remain still until a two clustered W event will show up.
Actually they couldn't suggest any bet whether the actual W/L count is in the positive or if the actual shoe was already played for the most part.

After all, if a given "more likely flow" should happen at most part of shoes, some shoes must somewhat balance such slight propensity, a sure sign not to bet a dime.

See you next week.

as.

Patterns exploitation

I've stressed one million times that there are no ways to beat the game by using progressions (unless the method provides a FB advantage) or simply by thinking we are able to get more wins than losses by a magic mental power.
Or that after P or PP or PPP or PPPP the best bet to make is wagering B. This is total worthless rattlesnake.sh.it tested and tested and tested and getting (obvious) negative results itlr.
 
What we should be really interested about is the application of some random walks that have demonstrated, beyond any shadow of doubt, that move a lot (say well more than math expected) around the 0 neutral value.
It happens that some r.w's are very slight shifted towards one side of the operations, a thing that can only improve our results.

Technically is just putting ourselves to decide what to do after a specific pattern (of given  quality and given quantity) will show up infinite times and then registering the W/L step movements getting far or getting close to the 0 neutral value. Or, naturally, featuring very slow or slow step movements in a way or another.

This procedure needs some time to be exploited and it's not a coincidence that many casinos will entice players to bet every hand where the random world assumes the maximum peak and where the math casino's edge will take its full value. And of course patterns need some hands to be formed.

Then a pattern must always be classified by its quality and average range of apparition per shoe, a thing tending to limit further the "unknown" world Al was correctly talking about.

See you later

as.