21 January 2020


Al Kotov  Creative Commons BY 4.0


This is a working paper. For simple MASH, go here. For planner MASH, go here.



Every decision will be reduced to the choice of one of two.

Diads such as dichotomy or binary opposition are at the core of how the universe and our DNA works. DNA computers make 2^n computations at the time PCs make n. And 2^n is basically n binary choices. Symmetries are everywhere, and the decision-making, temporal, goes on everywhere on the levels of molecules and up.

There is a consensus that the prime goal of humankind is to put the DNA of the humankind on the black hole at the center of our galaxy. And the DNA structure is one of the most fundamental formulas in our universe (check out vortex math, yes, vortex math and electron-Grover properties).

So we're here into something really huge definitely.



The GUI must be as simple as possible, and all the decisions are upon us the choice-makers.

We must have our own criterias for our choices. And computers are there to make calculations. The machinery must only provide the most convenient GUIs and the easiest systems to use.


A remark on simplicity: if we talk about anything in any ontologies or systems of coordinates & we have goals as areas there and we want to get there the quickest way possible, the path will strike for simplicity & will probably have zeros on some axes. That math is simple.


The first question of real consideration is how the choice between the two must affect the rating.


Elo rating system fits us well.

It works like this. With each decision, the winner gets some amount of score points from the loser. This amount is defined by the difference between their score points.

Chess and many online games have been using Elo for a long time.

It has numbers that are experimental and results tend to shift towards some locals. But for now, let's take the Elo.

It builds the score rating which can give us unexpected and useful insights and severely improves our decisions fast, and that's what we want.


First implementations

Let's describe them as they come to mind:


Simple MASH

If we have just one choicemaker and we want the most explicit rating, we can use a mergesort algorithm that will give us a minimum number of choices, but that's still a lot - for instance, 500 pairs for me and a set of 70 fragrances. In mergesort MASH new choices can affect the rating too much, so you wouldn't be able to skip a pair. And what if you'd want to reconsider?

Random choice of pairs is a better way to go, it will give us more insights sooner in the most cases & a lot of people can contribute.


So here’s how the simplest MASH we want works shortly:

We upload a set of items (images and/or text in current implementation) to make choices about. MASH shows us two randomly picked items. And we choose between them. And we can share the link to this MASH to whomever we trust to choose.

This way it makes sense to compare what’s comparable. Not choose between Lamborghini and a pair of glasses.


So what if we have

MASH with entities having inherent values

? These vales can be initial and changeable. That’ll be called RADICAL MASH or BANDIT MASH. We’ll talk about that later.

One great application for MASH with inherent values is financial planning, where we don’t care about when to buy. We do only care about an order in which to buy - as soon as we get an opportunity. The resulting rating will reflect that order.

With eight being the best number of items to make decisions upon, let’s pick two to fourteen items randomly, eight’s in the middle.

Our program will divide these in two groups with total values close to equal. And it will add some pure value (money or medium) where there’s less for the choice to be 100% fair.

If our our random picking makes no two groups that can be comparable because of the big difference in total values, the program will try to add more items picked randomly, – until it creates a pair with the total difference lower than, say, half of the value of the least valuable group in the pair. We must define this number for the program.

Our choice will affect all scores proportionate to inherent values of items in two sets, including that of money. Yes, we’ll know the value of money for what we’ve uploaded.

Instead of money we can use, say, other numerical mediums. Or we can measure the value we get not the value we give.


Bandit vs A/B

We have described the simplest cases where items are picked randomly. But what if our choices actually depict the interest we want to give items in MASH? Scores will then be aligned with the probability of items showing up in new pairs.

This is called bandit testing. Unlike A/B testing with all items picked randomly, which is commonly used on old and stable markets, bandit testing is normally used in short-term campaigns and on evolving markets, real-time.

Bandit testing can be an option for every MASH. We can use either bandit testing or random picking of items.

With the world and computations evolving, we’ll definitely put bandit testing in our equation.



What if we extrapolate bandit testing to values? Radical MASH, as in radical markets, is what we’ll get.

Choices will not only affect the scores or the probabilities of items showing up in MASH, they will affect the values.

We can start with changing values by clicks and add the choicemaker’s value-weight to that. This can be one number for one choicemaker, e.g. his free capital or his influence. Or this can differ from choice to choice depending on his confidence in the decision.


One way to think about it is in layers:

1. No values. Simple MASH

2. Ideal. Just clicks.

3. Ideal. Clicks multiplied by the value-weight of the choicemaker.

4. Ideal. Clicks multiplied by the value-weight of the choicemaker and his confidence in the decision

5. Real. Each choice changes values. The choicemaker becomes the marketmaker. These markets are called radical.


Changing values will fluctuate around scores of even the simplest MASH. Values are supposed to add initial impulses towards equilibrium.


This paper is a work-in-progress, but it plots the path for a lot of further research already, such as studies of users, user-defined parametrizations and classifications that are a job of companies by another consensus.

Please, share ruthlessly: