Roulette Martingale alternative (Carsch)

Here is a so called negative progression (up as you lose).

Note that I am not endorsing this or saying anyone should play a negative progression, you are guaranteed to pay more house edge raising bets.

However, if you are hell bent on using a negative progression like a martingale, or Fibonacci, etc perhaps take a look at the Carsch progression. It is for use with EC (even chance) bets only.

All the caveats of the martingale apply. A sustained run of losses can overpower it and whilst it takes a substantial 30 straight losses to bust, that is still no guarantee.

As with any progression that doesn't attempt a full recovery on a single win it is susceptible to repeated loss streaks prior to a full recovery.

Any two wins clears the progression. It allows you to catch streaks, and avoids bet escalation with WLWLWL patterns.

The progression has three levels and requires 558 units to play all three levels:

Base Level = 62 units
Recovery Level 1 = 124 units
Recovery Level 2 = 372 units​

Base Level

The progression is as follows (units):

1u, 1u, 2u, 2u, 4u, 4u, 8u, 8u, 16u, 16u​

How to play:

• 1u win & parlay (lose and move on to the next bet)
• 1u win & parlay (lose and move on to the next bet)
• 2u win & repeat the same bet (in order to move on to the next step, you must either lose this bet the first time, or win it and then lose it twice in a row)
• 2u win & parlay (lose and move on to the next bet)
• 4u win & repeat the same bet (in order to move on to the next step, you must either lose this bet the first time, or win it and then lose it twice in a row)
• 4u win & parlay (lose and move on to the next bet)
• 8u win & repeat the same bet (in order to move on to the next step, you must either lose this bet the first time, or win it and then lose it twice in a row)
• 8u win & parlay (lose and move on to the next bet)
• 16u win & repeat the same bet (in order to move on to the next step, you must either lose this bet the first time, or win it and then lose it twice in a row)
• 16u win & parlay (lose and move on to Recovery 1)

Recovery Level 1

This is played just like the base level but with the following unit sizes:

2u, 2u, 4u, 4u, 8u, 8u, 16u, 16u, 32u, 32u​

Drop back to base level when 62 units are recovered.

​

Recovery Level 2

This is also played just like the base level, with the following unit sizes:
6u, 6u, 12u, 12u, 24u, 24u, 48u, 48u, 96u, 96u

Drop back to Recovery Level 1 when 124 units are recovered.

You will need to keep track of your level and the recovery target when on a recovery level.

 

Does this progrseeion have a name?

 

I just realized it does. It's called the Carsch progression.

 

The obvious question is : what are your chances of losing? With this progression you need a parlay to win within 30 spins, so what are the chances it won't, and hence the progression busts?

It turns out there's an interesting relationship between the Fibonacci sequence and the probability of not getting at least two-in-a-row in a sequence of n trials for an even chance.

A formula for the Fibonacci sequence is F(n) = (Phi^n − phi^n) / ,

where Phi = (1 + )/2 and phi = (1 − )/2 are the roots of the quadratic equation x^2 − x − 1 = 0.

The probability that no two consecutive heads appear in n tosses of a coin is F(n+2) / 2n = (Phi^(n+2) − phi^(n+2) / 2^n· .

In this case n = 30. Using this online calculator, F(32) = 2178309 and 2^30 = 1073741824, so the probability is 2178309/1073741824 = 2.028708346 x 10^(-3) or about 1 in 493. This is the probability of a progression bust, so the inverse probability (that you will win) is about 99.8%. This is equivalent to a standard martingale of 9 steps. Obviously we're not dealing with a fair coin here and the probability is a lower for a casino game.

 

Typo, that should be F(n+2) / 2^n.

 

In my opinion;

rather than following a rigid standardized model .. it is always better to .. 'according to the hit resulting exposition' .. adjust the progression amounts to either .. from there .. 2 or 3 hits -- or more hits 2+3, 3+2 hits combos;

adjusted for whatever the protrusion rate to the new high is, ideally nominal .. thus self-incurring the lowest inposition increase rate(s) & thus, also self-incurred by design, method's volatility.

 

Nice analysis, thanks for that!

Something to also note the progression wins 3 units when the first bet+parlay wins. And win streaks will be capitalised on better than 1u due to the built in positive progression on the first win+parlay.

Being roughly equivalent to 9 step martingale with only a 512 unit drawdown is half the martingale bankroll requirement. However the major disadvantage vs the martingale is that the marty will clear all the debt in a single win.

I still wouldn't use it but I posted it for comparison purposes.

 

For those interested in a detailed analysis of the martingale, this is a solid paper (warning math ahead):

https://digitalscholarship.unlv.edu/cgi/viewcontent.cgi?article=1630&context=gaming_institute​

TL;DR don't do it.

 

That guy CARSCH was already a joke om GG. What bull sheet revival .Courtesy math zombie .


ROFL.

 

An interesting ancient thread ..
https://betselection.cc/index.php?topic=4997.0

 

The logic there is reasonably sound.

I've demonstrated this in building bankroll videos that taking profit within a statistical upswing can work.

Be modest aiming for 20% but be willing to accept less if things start to grind or you catch a long downswing from the start then be satisfied to get back to even. Use both a profit target and a session time limit. Session bankroll is always what you're willing to lose to catch the posisitive variance / statistical upswing but stick to the time or number of bets limit.

 

Spike checked out an rng that I use for testing and said it was too random. Is there such a thing as too random?

 

Quite the opposite to reality, but that is the incoherent babble we have to accept from him. What he calls "too random" may be his terminology for "different behaviour".

RNGs struggle to be random, it's literally impossible to be true random without real entropy. All RNGs attempt to stretch out a very small initial entropy state or seed. It is kind of like watering down a tin of paint and trying to cover the entire city. Sure you can do it but the thin veil of deception can be seen through with some tests.

One of the tests that they fail is generating blocks of data, think of each block as a large number or a pocket on a large roulette wheel with millions of numbers. Real random will have a certain collision rate on those numbers (think repeaters), whilst an RNG will not repeat as expected because they cannot both produce all possible numbers with equiprobability and also produce output that has the same repeater/collision rate as real random.

Obviously the statistical characteristics of the output will be different. But there are many other problems where internal state of the RNG can produce low quality output for sustained periods.

 

There is a way. You use the same technology used to make a physics computer for cheating. You just create a bunch of random entry points and have it land in a slot after it has passed random speeds, random bounces off of obstacles , and random skips off of slot barriers. It would be a real world physics based RNG.

 

Slots get their entropy from the instant a player presses a button. It's still pseudo random but good enough for the house.

Slot RNGs are deterministic using a fixed seed and they constantly generate a stream of numbers, thousands or millions per second and it's the precise timing of the button press that locks in a number.

Johnson Nyquist noise is also a much simpler and relaible mechanism than a mechanical process for obtaining entropy (that is how entropy is generated for high assurance cryptography devices):

https://en.m.wikipedia.org/wiki/Johnson–Nyquist_noise
​

 

That's good. I want anything better than common simple physics acceleration and obstacles for cryptography. But for a basic fair RNG a few distance hurtles and a random drop to kick it off would be good enough.

It does not matter. The casinos or R-Sim is not going to go beyond what's easy to program in basic modern function libraries and classes.

 

I appreciate the effort you put into answering not only my questions but everybody elses. But here are the real questions I am trying to get answered. If I consistantly get good results testing my strategy on rng's, does that mean I will not get good results from actual live play? Or, could I expect better results from live play? Is it harder to be successful with live play or harder to be successful using a rng. If I am simply testing a 50/50 outcome, what difference does a rng vs live play make? I have tested on every platform available and I see absolutely no difference. The patterns look identical. The results are the same. Could you clear this up so I will not be wasting my time one way or the other?

The only way I have been able to imagine what the difference between live and rng may be is the human element involved which influnces how the decision is arrived at. The speed the dealer gives the wheel. The speed and style he releases the ball, etc. Where as, a rng is more precise in it's selection, no human element. Even at that, will not patterns form in an identical manner between the two? XXOOXXXXXOOXOXXO. Was that pattern formed from live play or a rng?

I am constantly accused of being a kitchen table player. But so what, I would not be such a fool to play for real money unless I thouroughly tested my strategy first. But is my testing all in vain because I am testing it with something that produces results that are too random?
Thanks

 

I've suggested that fortuna would be a good start. I believe MacOS and iOS has adopted it because it was first introduced in the FreeBSD operating system on which they are based.
https://en.m.wikipedia.org/wiki/Fortuna_(PRNG)

Fortuna also assumes a threat model that includes poisoned entropy sources such as the NSA tampering of hardware random outputs from Intel processors.

Interestingly the Wikipedia page describes the phenomenon that RNGs face with statistical properties as per what I described previously around block collisions/repeaters.

With a 128-bit block cipher, this would produce statistically identifiable deviations from randomness; for instance, generating 2^64 genuinely random 128-bit blocks would produce on average about one pair of identical blocks, but there are no repeated blocks at all among the first 2^128 produced by a 128-bit cipher in counter mode.​

In the quest to have equiprobable distribution of outcomes RNGs can't both do that and produce repeaters/collisions at the rates that are expected and indeed present in random. Fortuna at least has an approach to manage this by rotating the encryption key from the entropy pool and is the only RNG that I am aware of that tackles this problem.

I still think there is room for improvement including XORing the output of multiple key streams with staggering of the key rotation from the entropy pools.

When it comes to ad-hoc things, a simple set of dice is hard to beat.

 

You may be getting invalid conclusions if the source of randomness is poor quality. I have witnessed RNGs on popular simulators get stuck in a phase where the quality of output is poor.

This can happen when the internal state of the RNG ends up with a lot of zeros. It happens and it is a flaw in RNGs and their ability to recover from internal states that have a lot of binary zero bits, it can take thousands of results to escape such states.

I can't say that is what is happening, but if you are using random.org entropy I would expect your findings if statistically significant would be similar.

Around 30-50k bets (not spins but actual bets) is a reasonable test to understand the characteristics and expectations of your method.

Live play is always harder. Distractions and mistakes and even dealer errors are part of the human factor. You know this playing craps that dealers make mistakes, players throw dice and you can struggle to get bets down in time. Dealers mantra is "A good games a quick game." Same with roulette spin rates as it impacts the casino hold and profitability.

Streak characteristics may be different because the EC is derived from either a spin outcome or a card shuffle. If the RNG has insufficient internal state to in principle emit all possible sequences during your session, or shuffle a deck (or 8 decks for baccarat) in every possible configuration equiprobably then you end up with results that superficially look random but actually fail statistical tests.

Most RNGs cannot perform a deck shuffle equiprobably so the possible candidate deck shuffles you get are a subset. They won't be able to provide all possible 1000 spins equiprobably either. They will try and distribute outcomes in the first order equiprobably but second order expectations fail. So whilst we expect all numbers to be uniformly distributed, the distribution of second order characteristics such as pairs of numbers may not be.

This gets back to the collision thing I have explained, faking random is a veil that if you look though too closely reveals some ugly flaws.

I would suggest this is because your tests are not statistically significant and you are not using methods to test the output.

Take a look at diehard tests, if you have spin results in a file you can test the outputs you have recorded.

Casinos always worry about collusion of dealers with palyers. Hence why tipping dealers is not permitted here.

Casinos monitor the results, that's how they pick up dealers who flash cards or signal players.

In practice I do not believe a dealer can target a number better than random because that would be an unacceptable risk to all casinos across the world. Controls would be used to prevent that, eg changing wheel speed and so on.

Targetting a number and using visual ballistics to estimate where a ball may land are two very different things. Again controls are used to make this increasingly more difficult.

Not if there are statistical anomalies.

Use data from real tables in addition to random.org and RNGs.

 

Patients of a saint twoup, that fool has been asking the same questions for years, various threads and answered with the same care and attention by many here , myself included. The information doesn’t get digested by him.

Thus.
I think he’s just an [removed, pay to advertise] codger wheelchair bound soliciting interpersonal exchange of any kind. Possibly a few Roos loose in the top paddock as well. Quite possibly martybating as he reads your replies, sick I know , but that’s Davie boyboy. Cheers.