車の一括買取査定申込み

車の一括買取査定申込み

Hybrid Intelligence for
Eective Asset Management
White Paper
Authors (Cindicator team):
Mike Brusov, Chief Executive Ocer and Co-founder
Yuri Lobyntsev, Chief Technology Ocer and Co-founder
Kate Kurbanova, Head of Analytics
Nodari Kolmakhidze, Chief Investment Ocer
Version 1.2.10
29 August 2017
Abstract
Cindicator creates the social and technological infrastructure needed to make
eective decisions under the volatile conditions of the new economy.
By combining a large number of diverse nancial analysts and a set of machinelearning
models into a single system, we are developing a Hybrid Intelligence infrastructure
for the ecient management of investors' capital in traditional nancial and

crypto-markets.
The benets of Hybrid Intelligence for an ecosystem and community are:
- a technological and analytical infrastructure for the ecient and safe management
of investors' capital by investors themselves or licensed managers;
- an opportunity for analysts to monetise their intellectual assets without
risking their own funds;
- tools and data for making investment decisions under the conditions of
market uncertainty;
- up-to-date analytics of the industry, expectations, opportunities, and market
growth points;
- indices and ratings of crypto-assets.
Contents
1 Introduction to Hybrid Intelligence 3
1.1 What is Hybrid Intelligence? . . . . . . . . . . . . . . . . . . . 3
1.2 Areas of application . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Venture investments . . . . . . . . . . . . . . . . . . . . 3

1.2.2 Science . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.3 Corporations and businesses . . . . . . . . . . . . . . . . 4
1.2.4 Politics . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Hybrid Intelligence for investments and asset
management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Ecosystem of Hybrid Intelligence 5
2.1 Collective intelligence . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Articial intelligence . . . . . . . . . . . . . . . . . . . . . . . . 7
3 Token sale 7
3.1 Expediency of issuing CND tokens . . . . . . . . . . . . . . . . 8
3.1.1 Eective economic motivation of all ecosystem participants 8
3.1.2 Necessity to business . . . . . . . . . . . . . . . . . . . . 8
3.2 Terms of token sale . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2.1 Terms of issue of CND tokens . . . . . . . . . . . . . . . 9
3.2.2 CND token distribution . . . . . . . . . . . . . . . . . . 9
3.2.3 Funding allocation . . . . . . . . . . . . . . . . . . . . . 9

4 An economic model for the ecosystem 10
4.1 Products for CND token holders . . . . . . . . . . . . . . . . . 10
4.2 Limited access to products . . . . . . . . . . . . . . . . . . . . . 10
4.3 Trading portfolio of Hybrid Intelligence . . . . . . . . . . . . . 11
4.4 Buyback of tokens for dynamic compensation
of forecasters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.5 Monetisation of intellectual contribution of
forecasters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.6 Technological infrastructure for investment funds . . . . . . . . 13
5 Technologies applied 13
5.1 Technological infrastructure . . . . . . . . . . . . . . . . . . . . 13
5.2 Data science and machine learning (ML) . . . . . . . . . . . . . 14
5.3 Description of current pipeline . . . . . . . . . . . . . . . . . . 15
5.3.1 Data ltration and clean-up (data preparation) . . . . . 15
5.3.2 Feature extraction . . . . . . . . . . . . . . . . . . . . . 15
5.3.3 Construction of hypotheses and mathematical models . 15

5.3.4 Validation and streamlining of predictive models . . . . 15
5.4 Description of validated hypotheses and approaches . . . . . . . 16
5.4.1 Conrmation of correlation between analysts' forecasts
and real market behavior . . . . . . . . . . . . . . . . . 16
5.4.2 Approach to the development of mathematical models . 17
5.5 Mathematical foundation . . . . . . . . . . . . . . . . . . . . . 17
5.5.1 Denitions . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.5.2 Superforecasting . . . . . . . . . . . . . . . . . . . . . . 18
5.5.3 Wisdom of the Crowd (WOC) . . . . . . . . . . . . . . 18
5.5.4 Model Boosting . . . . . . . . . . . . . . . . . . . . . . . 18
5.5.5 Sustainable Models . . . . . . . . . . . . . . . . . . . . . 19
5.6 Technologies (libraries, algorithms) . . . . . . . . . . . . . . . . 20
5.7 Technological roadmap . . . . . . . . . . . . . . . . . . . . . . . 20
6 Analytical products: completed and
in development 20
6.1 Binary probabilistic questions . . . . . . . . . . . . . . . . . . . 21

6.1.1 Macroeconomic events . . . . . . . . . . . . . . . . . . . 21
6.1.2 Corporate events . . . . . . . . . . . . . . . . . . . . . . 22
6.1.3 Political events . . . . . . . . . . . . . . . . . . . . . . . 23
6.2 Binary price-related questions . . . . . . . . . . . . . . . . . . . 23
6.3 Price-related questions . . . . . . . . . . . . . . . . . . . . . . . 25
6.4 Planned new data types, signals and indicators . . . . . . . . . 28
6.5 Planned new analytical products . . . . . . . . . . . . . . . . . 29
6.6 Public experiment with Moscow Exchange . . . . . . . . . . . . 30
1
7 Team and stages of development 31
7.1 Team . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
7.2 Current progress of the company . . . . . . . . . . . . . . . . . 32
8 Legal considerations 33
8.1 Legal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
8.2 Legal status of CND tokens . . . . . . . . . . . . . . . . . . . . 33
8.3 Legal status of crowdsourced forecasting platforms . . . . . . . 33

9 Conclusion 34
10 Risk factors and disclaimers 34
References 36
2
1 Introduction to Hybrid Intelligence
1.1 What is Hybrid Intelligence?
Hybrid Intelligence is the combination of human intelligence and machine intelligence,
and their interaction in resolving various tasks. One sort of intelligence
supplements and strengthens the other.
Clearly, one may face many challenges during the decision-making process.
Hybrid Intelligence and other related systems under development are appropriate
for resolving these kinds of diculties. This is not only due to the
criterion of speed in decision-making - namely, the question of why one should
waste time on simple tasks that can be resolved by both individuals and simple
mathematical methods and algorithms? It is also related to the complexity of
the tasks and the level of uncertainty in the systems used to resolve them.

In one of his latest interviews - Elon Musk speculates that humans should
soon merge with articial intelligence and create a new kind of interface.
This symbiosis could help people settle one of the most complex tasks facing
mankind: predicting the future with high accuracy.
People have long tried to resolve this issue in all areas of business by using
various technologies with varying degrees of success. Investors and traders try
to predict future share prices or company success to increase the protability
of investment deals. Political analysts try to predict the results of presidential
elections, while corporations put a great deal of resources into attempts to foresee
future technological trends. Many of them have already used intellectual
crowdsourcing to undertake these tasks to a greater or lesser extent. Let's take
a look at the existing solutions.
1.2 Areas of application
1.2.1 Venture investments
Most investment venture deals are closed by syndicates at the moment. This
means that several investors take part in one round of the deal at the same

time. This trend has been increasing year on year. In addition to syndicate
deals - which involve partner venture funds - specic associations and collective
investment clubs emerge each year (Angel-List is the most famous example).
Why do venture (and other) investors prefer group deals rather than individual
ones, despite the fact that sharing a protable investment with one's
competitors seems to make no sense?
One of the reasons for such deal structures lies in the use of collective intelligence
systems for risk hedging against the potential mistakes of group thinking.
This could happen when an investor makes the wrong decision about a deal
based on a false insight, trend, or insucient competence in a given area. In
a syndicate, a set of competencies and the investors' previous experience can
be very dierent, which allows them to view the startup as a whole, as well as
looking at the team and the potential risks from various angles and to cancel
the deal if there are sound reasons to do so. For most venture investments, the
best deal is a deal that has never been done before.
Now imagine the collective intelligence of professional investors combined

with articial intelligence technology, which, based on the use of a large volumes
of data in real time (including the number of exits, the stock market situation
in a specic area, the state of the labour market, and even the behaviour of
startup founders on social networks), adapts to current market conditions and
produces signals for entry or non-entry into a deal, free from emotional factors.
Most investors at the pre-seed and seed stages admit that emotions are still
the main drivers in investment decisions.
3
1.2.2 Science
The symbiosis of these two types of intelligence could, in this case, eciently
downplay the disadvantages of human 'emotional' approaches by strengthening
the decision-making signal with a number of decentralised data analysis
points. Using such a method is reasonable in systems with higher uncertainty
and highly complex of tasks, for example in biotechnologies. In a renowned scientic
paper, researchers created a game in which each player with a dierent
degree of knowledge could take part in molecular docking (a process that helps

predict the structure of a future chemical element with certain desired properties).
Each project participant could bind a molecule of the protein together
in any way. Using this crowdsourced data from a variety of experts combined
with virtual screening (computer modeling and machine learning) enables scientists
to create new medicines by combining a molecule (medicine) with a
target protein (cancer target). The synergy of the two types of intelligence
allows humanity to invent medicines for diseases that were once incurable.
In 1906, the famous British scientist Francis Galton came to a rural fair
where visitors were invited to guess the weight of a bull put on public display
and to write the gure on a special ticket, supposedly just for fun. The organizers
of that show promised prizes for those who managed to guess the correct
gure. Thus, about 800 people, some of them inveterate farmers, and others far
from pastoral aairs, took part in the voting. After collecting all the tickets for
analysis after the fair, Galton calculated the average arithmetic value from the
entire sample: 1197 pounds. The actual weight of the bull was 1198 pounds.
This means that as a group, the people gave an answer incredibly close to the

true gure.
As evidenced by statistics from the Who Wants To Be a Millionaire? television
game show, after contestants used the Phone-a-Friend option to phone an
erudite friend, a correct answer was chosen in only 65% of cases, but when the
player requested audience assistance, the aggregate answer from the audience
was true in 91% of cases.
Such studies of the wisdom of crowds were used widely amid the boom of
studies dedicated to group dynamics between the 1920s and 1960s. The sociologist
Heigl Knight, for example, asked students to estimate the temperature
in the room. The arithmetic average of the group's opinions estimated the
temperature in the room at some 22.44 ?N, while the actual temperature was
22.2 ?N.
1.2.3 Corporations and businesses
Google, Johnson & Johnson, and many other large corporations use 'collective
intelligence' for new corporate management technologies. Corporations have
already begun to integrate the technology of idea crowdsourcing and future

forecasting into their strategic processes to crowdsource new ideas and generate
forecasts for the future of the company and its competitors (sales plans,
new product releases, or entry into new markets).
Top management collects proposals and signals from various employees and
departments (on a decentralised basis, which is important), such as points of
view and 'insights' from sales managers, who collect market feedback on a daily
basis, as well as developers, who possess information on the actual value of technologies
and the company's fullment of its product plan, which are completely
dierent in nature and value. Combining this system with the unbiased algorithms
of big data processing (on sales data, analytical reports and forecasts,
and the constantly changing market situation) and mathematical modeling,
top management gains access to an extremely valuable decentralised source of
decision-making, which can be used in combination with other strategic parameters.
4
1.2.4 Politics
Certainly, a similar technology can be used for political purposes. A noteworthy

case is a well-known student project launched in 1988, the Iowa Electronic
Market. It turned out to be one of the most precise tools for predicting the
results of political events and elections for most countries around the world.
Participants in this 'market' can buy or sell contracts for the various results
of future political events (similar to short and long positions on the stock exchange),
thus forming expectations and determining the exact probability of
victory for one or another presidential candidate. For two decades, this technology
has been predicting the results of US presidential elections with great
precision, when compared with any analyst or company (until the most recent
election incidentally).
1.3 Hybrid Intelligence for investments and asset
management
Undoubtedly, stock exchanges are still the leaders in researching and using
Hybrid Intelligence in business - this is an area where traders have to make
decisions on millions of dollars every second (and trading robots do so every
millisecond).

Financial markets themselves are the daily prediction of the future in its
pure form. At what price and when is the best time to buy Facebook shares,
Brent crude oil, the US dollar, or Bitcoin? All these questions are the subject
of daily predictions of traders and analysts.
Major current market analytics related, for example, to forecasting the future
in terms of nances are created by a limited number of professionals using
roughly the same information. Each year the pace of information retrievals accelerates
and the value of such reports falls, with fewer and fewer professional
traders reading them or taking them seriously.
Nonetheless, such analytical reports bring in an impressive amount of money
all over the world. In 2015 alone, professional traders spent over $50 billion
purchasing nancial market data, of which $4 billion was spent on professional
analytical services and systems (predictive analytics). By 2020 this gure will
increase approximately six times. And these are only professional analytical
systems. The B2C nancial information market for non-professionals is huge:
for example, 54% of US residents have bought shares at least once in their lives,

and in China about 30% of residents are engaged in stock trading.
Recognising the potential and size of this market, we decided to put the
technology of Hybrid Intelligence to good use in nancial markets as a top
priority.
2 Ecosystem of Hybrid Intelligence
Thousands of analysts on the Cindicator platform generate various forecasts
daily, answering a number of specic questions about the price levels of dierent
nancial assets, macroeconomic indices, and events signicantly inuencing the
market.
Examples:
- create a forecast of the minimum and maximum price levels of Bitcoin
for the coming seven-day period;
- will the Tesla stock price surge to $345 during market hours on Friday?;
- will the U.S. unemployment rate be greater than or equal to 4.5%, according
to the 2 June report?;
5

- will Bancor collect more than 100 million during the rst week of ICO?;
- what is the probability of Trump's impeachment during the next three
months?
Cindicator works by using a large dataset that is transferred to a mathematical
block consisting of a machine learning model ensemble (cleaning, clustering
methods, linear regressions, Bayesian models, xgboost on decision trees, genetic
algorithms, and neural networks). Machine learning models dynamically calculate
various weightings for each forecaster, identify stable systematics in their
errors and calculate corrections for the errors, eliminate noise, and generate
nal predictions and trading signals.
At the core of our Hybrid Intelligence system is the synergy of the collective
intelligence of a large group of dissimilar decentralised analysts combined with
articial intelligence (machine learning, and a self-learning model based on a
variety of dynamic feedbacks).
Let's review these two ecosystem framework components in greater detail.
2.1 Collective intelligence

In order to ensure eective operations, any group intelligence system should
meet the following criteria.
The complexity of every goal set.
For results to have any relevance, they cannot be derived from extremely complicated
questions addressed to dierent users (i.e., the expected Bitcoin price
in USD in 376 days). However, the collected signal should present sucient
value. Aside from the complexity of the set task/issue, there is a need to create
the most convenient infrastructure for each participant to make such forecasts.
To do so, in December 2015 we launched a mobile platform where we focused
on the interface used by forecasters. As a result, it takes three to ve minutes
for each user to generate one data point.
Group diversity and decentralization.
Members of a single group intelligence should possess varied knowledge and
competencies, intelligence, personal experience, and views. If a particular segment
prevails in a group, the system will be incapable of generating an accurate
signal in the event of an incorrect insight.

A group may have a lot of outliers, errors, or subjectivity; however, the
diversity and multidirectionality of these points allows them to be ignored in
modulus (the simplest example is the Gaussian distribution).
Furthermore, the group should be completely decentralised. No communication
or exchange of opinions inside the group is allowed in order to avoid the
inuence of some individuals on others.
Motivation of each group member.
Each group intelligence member must be highly motivated to generate the most
accurate forecast possible (according to current knowledge and possibilities).
We have developed multi-level motivation for our platform, based on important
aspects of human psychology.
Financial motivation.
Each month, we distribute funds proportionately to each user's ranking in the
application. The more accurate the forecasts a user makes, the more compensation
they receive. Erroneous forecasts, and low activity downgrade the
6

rankings. Accordingly, each user's compensation depends on their personal activities
and the accuracy of the forecasts.
Competitive motivation.
We have developed internal user rankings, special nominations, and other gamication
elements to enhance the competition factor.
Involvement with trades and investment.
On our platform, users do not merely forecast in order to maximise their points;
each forecast is a micro-involvement of every user in a real or simulated trading
transaction or investment. Our trading robots complete a real or model
trade linked to every question posed. This is a signicantly promotion of the
involvement of every participant, both individuals and the group as a whole,
and increases responsibility.
Training.
Getting daily feedback on the accuracy of their forecasts as well as increasing
their level of knowledge before preparing each prediction helps forecasters to
enhance their skills and nd the best strategies for forecasting various types of

events.
2.2 Articial intelligence
The articial intelligence system is only the rst stage which generates a large
amount of 'raw' data. Next, Cindicator's 'black box' is used, with the following
core elements:
(1) the system and methods dening the condence weight (with constant
adjustments after each question and trade) for each user, which takes
into account:
- the personal track record of each member's accuracy, divided into
clusters (signal types, instrument types, links between answers, etc.);
- dynamic feedback following each trade with regard to the value
(prot or loss) of each user's forecast;
- the predictive model, which (in a very short time) is capable of
dening superforecasters in the group.
(2) trading strategies and models to seek the best possible way of using the
enriched data to create trading robots:

- testing of various trading strategies and hypotheses;
- constant backtests and forward tests to adapt the models to the
constantly changing market environment.
3 Token sale
By releasing CND infrastructure tokens, we oer all participants (traders, investors,
forecasters, analysts, data scientists, and the Cindicator team) the
chance to become the creators of a decentralised ecosystem of Hybrid Intelligence
for more ecient asset management.
Each CND token holder can obtain a new level of access to Cindicator's
indicators, indices, data, services, information, and analytical products. The
level of access and the products and tools available will depend on the quantity
of tokens in each holder's possession, which will in turn be inuenced by each
token holder's role and active participation in the decentralised ecosystem.
We also plan to place CND tokens on exchanges, giving people the opportunity
to buy them openly (for residents of countries where the purchase of
7

tokens does not violate local laws), gain access to new products, or sell them
to interested traders, analysts, or investment funds. Tokens cannot be sold to
residents of the USA, Singapore, or other countries where the sale of tokens
may require registration as a security.
We will make every eort to place CND tokens on the major cryptoexchanges
in the shortest possible time after the completion of the crowdsale.
However, legislation on the circulation of securities in certain countries, such
as the USA and Singapore, prohibits the sale of CND tokens to the residents
of those countries. When you buy CND tokens, you should be aware of the
restrictions on their subsequent sale and promise to follow our instructions
and/or those of the exchange when reselling them to other users.
3.1 Expediency of issuing CND tokens
The issuance of our own infrastructure tokens is conditioned by the need to
create an internal economy in the ecosystem that will establish transparent
and fair relations among all participants comprising the system: forecasters
and analysts, traders, nancial investors, data scientists, and the Cindicator

team.
3.1.1 Eective economic motivation of all ecosystem participants
Blockchain, decentralization, and a fair system of economic motivation are ideologically
and systematically integrated into the structure of the predictive
product module. Their purpose is to create a system of long-term motivation
that encourages forecasters to perform their intellectual work better, thereby
increasing the eectiveness of the entire technology and its benets to the community.
To ensure more ecient and fair motivation for active participants of the
ecosystem, (forecasters currently, but also data scientists and traders in the
future) we will locate direct causality between the quality of their engagement
and the result of real or simulated trading transactions or investments (which
are based on participants' forecasts, intellectual work, data processing models
and trading strategies).
For this reason, after the crowdsale and acquisition of the necessary licenses,
we will allocate part of the funding to the trading portfolio (managed by Hybrid
Intelligence). Potential prot from this portfolio will be used to buy back

CND tokens from exchanges (from their owners at market price) and the further
distribution of redeemed tokens between forecasters (in proportion to their
rating, accuracy, and participation over a given period). Therefore, nancial
compensation for active ecosystem participants will be directly linked to the
trading module performance.
3.1.2 Necessity to business
Throughout 2016 and early 2017, we launched test integration with hedge funds
and banks to monetise our technology by providing them with products and
APIs. We identied the poor scalability of this classic B2B modellarge funds
wanted to monopolise our technology, data, and trading signals (primarily because
of the limited market capacity  funds with the same valuable alpha
began to compete in utilising these signals). In other words, we realised that
selling our solution to a large number of B2B customers would be unwise from
a business point of view.
The issuing of infrastructure tokens is the next step towards the creation of
technological infrastructure (API + forecasting module + data science module

+ trading module + GUI module), which will be used by investment funds
working under the new format for utilising all products and capacities of Hybrid
Intelligence with maximum eciency.
8
The funds that will buy this technology (in addition to buying tokens on
the exchanges from their owners at market price) will pay a percentage (performance
fee) of their potential prots on a regular basis. Those payments will
be used for buying back tokens from the market, in order to increase motivation
for all active participants of the ecosystem (forecasters, traders, data
scientists). This infrastructure is scheduled to be available for funds in 2019.
In order to preserve maximum eciency in the utilisation our technology
by investment funds (hedge funds, crypto asset funds, venture capital funds)
access to this infrastructure will be granted only for those who own a signicant
number of CND tokens.
3.2 Terms of token sale
3.2.1 Terms of issue of CND tokens

CND tokens will be issued on Ethereum blockchain using the ERC20 token
standard.
Token sale period: 12 September to 12 October.
Prior to the crowdsale, we plan to start selling the tokens via the White List
in several iterations. There is a possibility that all tokens will be sold through
these stages before the start of the crowdsale.
100% of the tokens will be issued within the token sale period.
Purchase methods accepted: ETH.
Price of 1 CND = $0.01 (at price equivalent provided for illustrative purposes
only, no at currency will be accepted).
Maximum hard cap = $15,000,000.
3.2.2 CND token distribution
Tokens will be distributed as follows:
75%  for token sale contributors;
20%  for the Cindicator company (+ vesting);
3.8%  for advisors and partners;

1%  for the bounty campaign;
0.2%  for current Cindicator forecasters (proportionally to their total
rating).
3.2.3 Funding allocation
Funds will be allocated as follows (proportions below are not nal and may
change at company's discretion based on business needs):
55% - budget for continuation of scientic work, infrastructure development,
creation of new products, development of a Hybrid Intelligence
platform. The budget will be allocated between these areas as set out
below:
- development: data science, machine learning, AI modules, mobile
applications, web versions, products, API, web-hosting, server capacity;
- trading: trading services and terminals, development of trading algorithms
and infrastructure;
- operational costs: salaries, oce rent, other operational costs.
9

20%  Hybrid Intelligence portfolio for technology validation, the accumulation
of valid trading data and formation of a dynamic motivational
fund for forecasters. The trading cases of this portfolio will also serve
to make up a history of transactions, which will contribute to growing
interest and demand for Cindicator products in the professional market
of investors and traders.
10%  marketing: promotion of the collective intelligence platform in
order to achieve signicant growth in forecaster numbers.
5%  legal support, improvement of company's legal structure, protection
of investors' rights.
5%  monthly forecaster compensation fund.
5%  acquisitions and future partnerships for the synergetic development
of the Hybrid Intelligence ecosystem.
4 An economic model for the ecosystem
4.1 Products for CND token holders
By buying tokens, CND token holders will get exclusive access to part of the

Hybrid Intelligence infrastructure now under development.
Holders of CND infrastructure tokens will receive a dierent level of access
to Cindicator's indicators, ratings, and internal analytical products.
Token holders will be able to access the following parts of the infrastructure:
- indicators of traditional markets and crypto-markets (the probability of
the rise or fall of asset prices, the probability of beating consensus in corporate
and macroeconomic events, indicators of certain price levels being
reached, and indicators of the probability of signicant events inuencing
the market);
- auxiliary service products for trading (Telegram and Slack bots, notiers,
and portfolio monitoring products);
- analytical products (ICO ratings, market condition analysis, ICO due
diligence, and investor portfolio analysis);
- market indices and sentiments generated by Hybrid Intelligence.
The fact that token holders can use data from the analytical infrastructure
products will not aect the value of the data received from Hybrid Intelligence,

since each indicator or index is not an unambiguous trading signal, but only
an additional metric in the market that helps analyse an investment decision.
These data and analytical products will assist token holders and make the
ecosystem transparent.
However, a part of the infrastructure intended to be directly used in capital
management (by traders' teams, machine learning models, and trading strategies)
will remain in the centralised part of the system. This is necessary in
order to make sure that Hybrid Intelligence can be used most eciently at the
next stage, when interested funds will be provided with access to the entire
infrastructure (see Section 4.6).
4.2 Limited access to products
To prevent the dilution of analytical data value (taking into account market
capacity and the potential impact on it), the access level and set of available
products and tools will be provided for CND tokens.
The precise formation of the levels will depend on the results of the crowd
10

sale (the quantity of tokens released). We will publish the rst version of the
access grid for our products, signals, and data after the completion of the crowd
sale.
We will deliver these products in various ways once the corresponding development
work is complete:
- Daily/weekly/monthly distribution of indicators via messenger/email;
- SaaS(Software as a Service)  a website with an access to indicators and
analytics of Hybrid Intelligence for various events;
- Mobile application;
- API access.
4.3 Trading portfolio of Hybrid Intelligence
The funds that will be used to create a dynamic motivation fund for forecasters
will be divided into three parts in order to cover the most protable and scalable
trading strategies, as well as for the eective hedging of risk:
1. Active cryptotrading based on Cindicator technologies, along with data
and signals retrieved from the consensus of Hybrid Intelligence. This

portion of the portfolio will vary with cryptomarket liquidity. At the
moment, liquidity makes it possible to comfortably use a small percentage
of the total portfolio in active strategies. The result of trading this part
of the portfolio will be quantied based on the Bitcoin as a benchmark.
2. Protective buy and hold portfolio of crypto assets (the proportion of
various assets in the portfolio is determined by consensus). The task of
Hybrid Intelligence is to determine the optimum ratio of crypto assets
from the viewpoint of risk minimisation, and to keep this ratio up to
date. Bitcoin will be used as a benchmark to estimate the results.
3. Active trading of traditional nancial assets: stocks, futures, and foreign
exchange markets on the basis of Cindicator technologies, as well as data
and signals retrieved from the consensus of Hybrid Intelligence. This
part of the portfolio is used to demonstrate the capabilities of Hybrid
Intelligence in traditional markets. It can also be treated as protective
in relation to the entire cryptoportfolio. In the case of a strong fall in
the cryptomarket, a portion of the funds may be transferred to cryptoassets

with the purpose of earning a prot upon the rehabilitation of that
market. USD will be used as a benchmark to estimate the results.
Active management of the third part of the portfolio will begin within a few
months after the crowdsale is completed. To do this, we will need to complete
the preparation of the entire trading infrastructure, such as accounts and legal
structure (it is necessary to establish a separate legal entity for the fund and
to obtain the necessary licenses).
This portfolio will be managed by our team of traders and trading robots,
who will use the data, signals, and analytics obtained through the Hybrid Intelligence
technology. We will apply various strategies in the nancial markets
(both crypto and traditional) within dierent time horizons, from short-term
trades to long-term investments. The choice of strategy and assets invested
will draw on a positive evaluation from Hybrid Intelligence, as well as successful
testing in the form of back and forward tests.
Our team will prepare detailed monthly reports featuring the results of the
trades executed and make them available to the community.

11
4.4 Buyback of tokens for dynamic compensation
of forecasters
Every quarter we will record the results of all the accounts in the Hybrid Intelligence
portfolio in order to form a CND pool, designated to reward forecasters
for their intellectual investment into the ecosystem. Buyback of the tokens
(in case of positive performance) will be executed as following: tokens will be
bought from holders at the price set on exchanges, thereby ensuring the liquidity
of CND.
In the case of positive performance (relative to the initial state of the portfolio),
we will distribute the prot as follows:
1. X% will remain in the Hybrid Intelligence portfolio, ensuring its growth
for the next reporting period;
2. Y% will go towards buyout of tokens on exchanges at market price, which
will then be distributed as additional nancial motivation among the forecasters,
forming consensus in proportion to their ranking and contribution

to the positive result of the portfolio for the given reporting period;
3. Z% is the performance fee for the Cindicator team (to be paid only if the
portfolio size is larger than its initial state).
In case of a loss, we will use the reserve fund to buy tokens from exchanges
and provide nancial motivation and compensation to the superforecasters for
the period in question.
The buyback will be done on a step-by-step basis, between two scal quarters
to avoid the risk of unfair market manipulation; Cindicator will make
appropriate reports available to the community on completion of quarterly
buyouts.
The main purpose of buying back tokens is to ensure CND circulation for
motivating forecasters to conduct market analysis of even higher quality and
to increase the number of precise predictions.
The parties acknowledge that the current state of the cryptocurrency industry
is uncertain due to fast changing regulations and/or lack of regulatory
certainty in several jurisdictions. To comply with any regulations and/or to

ensure viability of its business model in light of any market, technological,
and/or regulatory changes, Company reserves the right to amend, supplement,
or delete any term of this Agreement, including but not limited to any terms
dealing with the potential buy-back of CND Tokens.
4.5 Monetisation of intellectual contribution of
forecasters
Forecasters forming the Cindicator collective intelligence are the key element
to be created with in the ecosystem. The fruitful operation of this system as
a whole requires the personal motivation of forecasters to be sustained, and a
common goal to be formulated for the entire group.
Our platform enables professional and non-professional analysts to monetize
their intellectual work in analysing markets and generating predictions. We
call this product the Collective Intelligence Platform, in which our forecasters
can invest their mental asset (time, attention, intelligence) and be eligible for
the respective compensation of their embedded intellectual investments, with
no risk of losing their own nancial assets.

Personal motivation
Each forecaster generating various forecasts in our application is given a
personal rating, based on the forecasts' accuracy. The rating may both increase
and decrease, depending on the accuracy of each prediction. The rating
12
of each forecaster is made public, which creates the necessary competitive motivation
for each of forecaster.
At the end of each month the rating is xed, and the top 2% of forecasters in
the ratings share the cash prize in proportion to the number of points accrued
that month. The monthly cash prize is formed from the reserve fund of forecasters'
remuneration and depends on the number of forecasters at the time.
The size of the monthly prize and the rules for its allocation are announced
before the start of each month. Within approximately one month we will buy
tokens back on exchanges in an amount equal to this monthly prize in order
to pay the CND tokens to forecasters for their intellectual contributions at the
end of the month.

In the beginning of each month, this rating is reset (in order for everyone
to have an equal chance in the new period) and a new monthly stage begins.
Group motivation
The overall goal of the entire forecaster group will be related to the result
of trading the Hybrid Intelligence portfolio, since they are an integral part of
its management.
At the end of each reporting period (quarter) we will record the results
of trading in the Hybrid Intelligence portfolio (for the traditional portfolio, in
USD; for the crypto-portfolio, in BTC). If prots are gained on any of the accounts
(according to the corresponding benchmark), a part of the prot will go
towards buying back tokens from exchanges, which we will allocate to forecasters
as an additional bonus in proportion to their rating in the given reporting
period.
In case of a loss on both accounts, no additional bonus will be provided to
the forecasters.
4.6 Technological infrastructure for investment funds

The nal goal is to create a complete infrastructure for the new generation
investment funds which will buy access to Cindicator technology (API, forecasting
module, data science module, trading module, GUI module, security
system).
The funds will be able to access this technology by purchasing the necessary
numbers of CND tokens (see section 4.2) on the exchanges (from their
owners at market price). The funds will also purchase a license by paying a
percentage (performance fee) of their nal portfolio. Those payments will be
used for buying back tokens from the market, in order to increase motivation
for forecasters and other active participants of the ecosystem.
The number of funds that will get access to the full infrastructure will be
limited in order to maximise the eciency of the infrastructure on each market
(cryptoassets market, traditional stock market, currencies market, derivatives
market, venture capital market).
This will provide an ecient supplement to Cindicator's ecosystem, increasing
its sustainability and providing benets for all active members.

5 Technologies applied
5.1 Technological infrastructure
The Cindicator technological infrastructure is already developed at the time of
the token sale and consists of the following modules.
Business logic module:
13
- Backend system with basic business logic that works with events;
- Administrative system;
- Viewing data and indicators;
- Mobile applications (iOS + Android);
- Web application (under development).
Prediction module:
- Data acquisition;
- Filtration and cleaning of acquired data;
- Feature extraction;
- Forming of hypotheses and mathematical models;

- Validation and optimisation of parameters for predictive models;
- Synthesis of accurate predictions.
Trading module:
- Data acquisition from the predictive module;
- Integration with exchanges, acquisition and processing of resulting data;
- Back tests and forward tests for parameters of trading strategies;
- Implementation of trading strategies through trading robots.
5.2 Data science and machine learning (ML)
ML is employed by Cindicator to accurately forecast the actual behaviour of
nancial instruments based on data from the market and forecasters' predictions.
To achieve this goal, two major approaches are used: superforecasting and
the wisdom of the crowd.
We undertake this work in several ways:
1. We study our forecasters, identifying behavioural patterns and common
factors.
- We cluster forecasters: into bears or bulls, those who narrow or

expand price levels, analyse the market or not, follow the trend or
not, use technical or fundamental analysis etc;
- We explore behavioural patterns: how often forecasters make mistakes,
in which situations they are mistaken, and how forecasters
react to a dramatic change in the market and dierent economic
events.
2. We conduct experiments with groups and clusters.
3. We conduct experiments with predictive models and use them to build
the boosting algorithm.
4. We conduct time series analysis of the market and the predictions of
forecasters.
5. We validate machine learning models and optimise their parameters.
14
5.3 Description of current pipeline
Data available:
- Forecaster prole (gender, age, country, professional background, occupation,

and behavioural patterns);
- Forecasters' predictions for dierent nancial assets (binary questions,
price-related questions);
- Historical market data on various nancial assets.
We use the classical pipeline of machine learning models.
5.3.1 Data ltration and clean-up (data preparation)
The main source of random errors is forecaster input errors (where the user indicated
the wrong ticker symbol or specied an incorrect number order). These
errors adversely aect the work of models and displace our metrics. For data
clean-up we use the following methods: IQR, Grubbs Test, and GESD.
5.3.2 Feature extraction
Each forecaster and investment instrument has a distinctive behavioural pattern.
Our algorithms consider these patterns and apply either charge dierent
weightings or dierent models in the appropriate manner. We have developed
a model that constantly updates the feature vector and recalculates its weightings
based on RL (reinforcement learning).

5.3.3 Construction of hypotheses and mathematical models
All our models can be divided into two classes:
- Superforecasting models (in which we build models on various forecasters'
clusters and cluster ensembles);
- The wisdom of the crowd model (in which we build various models on
the predictions of all forecasters).
We develop mathematical models for description and prediction based on
the theories of phase transitions and game theory. We also use fractal geometry
to forecast critical points (points where the market experiences increased tension,
when the topological dimension and the Hausdor Dimension are changing
dramatically).
5.3.4 Validation and streamlining of predictive models
Our models are optimised and back test-assisted due to the pipelines involved.
Dierent models demonstrate their own specic behaviors for dierent investment
instruments. Each model has its own settings (the length of the sliding
window, the form of the function for calculating weights or penalties, the depth

of the decision tree, and others). Tuning of parameters is done for each model
with regards to each nancial asset. Each model is constantly learning on the
basis of new data. To assess the accuracy and quality of our models, we perform
back-testing and use both standard scores (RMSE, ROC, MAE, Pearson's
correlation coecient) and their intrinsic evaluation functions for each trading
strategy.
15
5.4 Description of validated hypotheses and approaches
5.4.1 Conrmation of correlation between analysts' forecasts and
real market behavior
To demonstrate existence of strong correlation between analysts predictions
and real market behaviour, we turn to the basic mathematical statistics.
Let R2  be the coecient of determination, one of the general mathematical
metrics, which determines the degree of correlation between the data. It
is believed that, when the condition R2 > 0.5 is satised, there is a strong
correlation between the data sets.

To calculate the coecient of determination for nancial instruments (GOOG,
BAC, SPY, MCD, Spc1). Let's demonstrate the values of the coecient of
determination on the example of data from the archive (Cindicator analyst's
forecasts and real values), calculated using an open library in Python:
from sklearn.metrics import r2_score
The obtained value of R2
:
GOOG: 0.8;
BAC: 0.93;
SPY: 0.94;
MCD: 0.9;
Spc1: 0.96.
For this calculations based on the data from data archive (forecasters predictions
reality) used an open library on Python language: from sklearn.metrics
import r2_score.
Conclusion:

Our experiments show the presence of a constant and strong correlation between
analysts' forecasts and real market behavior. This means that we can
implement a mathematical model that will extract from the forecasts of analysts
the exact values of market behavior with the necessary accuracy.
16
5.4.2 Approach to the development of mathematical models
Predictions' accuracy of each forecaster varies depending on the type of question:
some users provide great answers for questions related to macroeconomics
and politics, while making mistakes in price-related questions; others,
on the other hand, can accurately forecast price of a particular asset for next
day/week/month, yet make mistakes in questions of other types. By researching
behavior of each user we created self-learning system for assessment. For
instance, depending on forecaster answers for price-related questions we can
dene 6 dierent forecaster proles.
Wherein, behavior of the individual forecaster may vary: they predict only
trends on weekdays (no time to look at the charts), but make more accurate

forecasts on weekends. By applying dierent conversion operators, we can signicantly
improve prediction accuracy. In our models we take into account all
behavioral patterns of the users.
We have found that there was a group of forecasters who are especially good
at predicting the so-called bifurcation points, when the trend changes dramatically
or experiences a sharp jump/fall. By constructing a separate algorithm
for this cluster of forecasters, we can predict such bifurcation points and apply
individual models to them.
Our models use dierent ML/DL approaches, such as:
- A Bayesian approach;
- Bayesian Belief Networks;
- HMM;
- Using various models as separate predictors which serve to build up the
boosting;
- Building various regression models;
- Using various algorithms of clustering for segmentation and aggregation

of forecasters. We compose clusters of superforecasters and ensembles of
clusters, on which we run various algorithms;
- Using historical data on investment instruments in addition to user signals.
Models based on time series analysis are also used.
5.5 Mathematical foundation
5.5.1 Denitions
We dene:
? as a feature vector which characterizes a user;
Ui(?) as an i
th system user (forecaster);
U = {U1 . . . Un} as a user vector;
E = {E1 . . . Em} as a event vector;
Ereal = {Ereal
1
. . . Ereal
m } as real values (correct answers) vector for

events E;
Vl  as an l
th ticker or E-belonging event voting operator.
Then, a pure signal for event l, not involving algorithms, is denoted by the
following:
Sl =
Pw
i=1
Vl(UiEl)
w
=
Vl(
Pw
i=1
UiEl)
w

,
where w is the number of forecasts.
17
5.5.2 Superforecasting
The group Ul = {Ul1
. . . Ulm} is called superforecasters for events Ek = {Ek1
. . . Ekn },
if
Ul =U |Vk(UET
k
) ? E
real
k
|.
The value 
s

k = max |V (UlET
k
) ? Ereal
k
| is the limit error for the group Ul
at events Ek.
5.5.3 Wisdom of the Crowd (WOC)
We dene:
W as the user weights matrix;
Wl as the user weight vector for the event l;
WT
l U as the weighed users vector for the event l.
Then, the WOC algorithm signal is denoted by the following:
Sl = Vl(WT
l UET
l

).
The value

w
k = max[Vl(WT
l UET
l
) ? E
real
l
|
is the limit error of the WOC algorithm at events El
.
5.5.4 Model Boosting
We dene:
μ

1 = {μ
1
1
. . . μ1
n} as the Superforcasting models (forecasts) vector;
μ
2 = {μ
2
1
. . . μ2
n} as the WOC models (forecasts) vector.
The following linear combination of models is called model boosting:
L =
Xαiμ
1
i +

Xβiμ
2
i
.
In matrix form:
L = AM,
where A = αiβj is the coecient (weight) matrix, M = μ
1
i μ
2
j
is the model
matrix μ
1 and μ
2
.

We shall select a matrix A?
, that
A? =A |AM ? E
real|.
The limit error of model boosting will be the following:

b = max |AM? ? E
real|.
Therefore, the following condition will be correct:

b <= min(
s
, w).
18
5.5.5 Sustainable Models
Let us consider the class of models μ?(, τ ) = {μ?1 . . . μ?n|μ?i ∈ μ

1 ∪μ
2}, for which
the following sustainability conditions are laid:
1) ∀t ∈ T
τ
l
d(E[E
predict
t ? Ereal
t
])
dt = 0, i.e. expectation of forecasts is a
constant;
2) Var[E
predict
t ? Ereal

t
] < , i.e. error distribution of expectation does not
exceed ;
3) Var[E
predict
t ? Ereal
t
] < E[E
predict
t ? Ereal
t
], i.e. forecast distribution does
not exceed expectation,
where Epredict is the forecasted value (for any algorithm), T
τ
l = {tk1

. . . tkτ } is
the time series of the ticker l.
Such a class is called sustainable under the timing error.
Theorem 1. Let us assume that μ?(, τ ) is the model class sustainable against
the timing error, T
η
l
is the time series for which T
η
l > Tτ
l Then:
∃ χ : S → S is the ane transformation at a set of signals, for which:
limt→∞
|S?
l ? Ereal
l

| = 0, aaa S?
l = χ(Sl), t ∈ T
η
l
.
Proof. Let us prove that provided that Theorem conditions are met, there exists
an ane transformation χ, for which |S?
l ? Ereal
l
| &.
Without prejudice to communality, let us assume that all the forecasts are
shifted upward, beyond the real value by γ ?  (inferred from the denition of
a sustainable class). The case of downward shifting is proven analogically.
1) Induction Basis. Let us assume that γ <  is the forecast error for the
moment t1 = T
τ

l
, then let us set the shift χ(Sl) = Sl ? γ as the transformation.
We might note that the error for this case will be γ? = 0.
We shall add the forecast with the error α <  to the moment t2 = T
η
l
.
Using the previous transformation, we then obtain that the error of the
second forecast became as follows: α? = α ? γ.
We then see that without the transformation, the total error of two forecasts
will be γ + α, and α ? γ with the transformation.
It is obvious that α ? γ ? γ + α, as γ ? 0.
2) Induction Step. Let us assume that γτ = {γ1 . . . γτ |γi < } denotes errors
of forecasts in the time series T
τ
l = {tl1

. . . tlτ
}. Then the value γ =
P
i
γi
will be the total error in the series T
τ
l
.
We shall take the transformation χ(Sl) = Sl ? E[E
predict
t ? Ereal
t
], where
t ∈ T
τ

l
.
Let us then add the next forecast with the error α <  to the moment
tl(tau+1) = T
η
l
.
Now we must prove that
X
i
γi + α ? (
X
i
γi + α) ? 2
P
i

γi
τ
,
or
2
P
i
γi
τ
? 0.
19
As τ 6= 0, while the sustainability conditions provide for P
i
γi ? 0, then the
monotonicity statement will be correct.
Let us use the Weierstrass Theorem for the limited monotonic sequence:

Any monotonically increasing series, bounded from above (or
monotonically decreasing series, bounded from below), has the
limit equal to its least upper (lower) bound.
Thus, we obtain that limt→∞
|S?
l ? Ereal
l
| = 0.
All the models used in Cindicator hybrid intelligence technology, could be
abstractly set out in the form of approaches given above. On account of the
described theorem, there exists a model transformation for which error expectation
reduces to zero in the time series.
5.6 Technologies (libraries, algorithms)
- Languages: Python, Scala, R;
- Libraries: NumPy, SciPy, Pandas, scikit-learn, matplotlib, seaborn,
keras, Theano, xgboost;

- Algorithms: regressions, clusterisations, ARIMA, boosting, decision
trees, random forest, deep learning;
- Infrastructure: Django/Flask/Tornado, Postgres, MongoDB, Re-dis,
MS Azure, Hadoop, Spark.
5.7 Technological roadmap
In the future, as our technology develops and amount of data increases, we
plan to:
- Implement neural networks and deep learning;
- Implement a trading robot based on reinforcement learning, which will
independently analyse the market and learn from its own mistakes;
- Develop modern mathematical models to build predictive models for the
market;
- Collaborate and cooperate with data scientists from leading universities
(Stanford, Berkeley, Princeton, SPSU) and companies (Google Research,
IBM) in nance, data science, and ML/DL;
- Create a platform for managing trading robots;

- Develop the market2vec algorithm (a vector representation of nancial
assets' data).
We believe that the merging of such areas as control dynamics, game theory
and technical analysis, machine learning, and behavioural analysis is a very
promising eld.
6 Analytical products: completed and
in development
Over 8,000 forecasters have made 230,000 forecasts since the global platform
launch in December 2015. In July 2016 (once an adequate data set had been
accumulated), the company started trading, forward testing and backtesting
various trading strategies.
The main focus areas included two types of questions:
20
- binary probabilistic questions;
- price-related questions.
6.1 Binary probabilistic questions

Binary probabilistic questions are questions that only have two possible answers:
yes or no. Forecasters have to give their answer as a 0%100% probability
for an event to happen. A 0%-49% probability is interpreted as a no
answer with various degrees of condence, while a 51%100% probability is
interpreted as yes with various degrees of condence. As a rule, this type
of question is used to forecast political, macroeconomic, corporate, and other
sorts of events, as well as to forecast price movements to a certain degree.
Each analyst uses various strategies to answer these questions, as a specic
probability value set out by a forecaster aects the number of points in the
forecaster's rating*that is, their nancial motivation. Thus, each user is both
an analyst and a risk manager for their portfolio on the platform.
For instance, some forecasters choose 40% or 60% for most questions; therefore,
they can lose (if the answer is wrong) or gain (if the answer is right) 10
points maximum. Some forecasters only choose 0% or 100% to receive the
greatest possible number of points (similar to traders, who use an equal risk
per position in each trade). And some forecasters are cautious in events of

great uncertainty and aggressive (0%10% or 90%100%) in events where they
are more condent. Therefore, they are similar to traders, who risk bigger with
greater condence in a trade, and vice versa.
As a result, articial intelligence models weigh the answers and assess forecasters
over the history of their forecasts, as well as forming dynamically changing
clusters (bulls, sheep, bears, superforecasters, etc). As a result, we get a
valuable aggregate signal, which can be used in various trading systems and
strategies.
Let's take a closer look at various types of binary probabilistic questions.
*The number of points is calculated as the dierence between 50% and the
user's answer. For instance, if a user chose 65% probability for the event Will
the Bitcoin price rise to $3,500 by July 30? and the correct answer is no,
(i.e., 0%), they will receive 50  65 = -15 points. The forecaster who answers
10% for this question will get 50  10 = 40 points for their answer. Therefore,
the maximum number of points available is 50. For more information, please
see our FAQ.

6.1.1 Macroeconomic events
Macroeconomic events are the latest and most important economic events,
gures, and facts that could aect nancial markets. For example, we needed
to nd out the probability for the Federal Reserve's interest rate hike in June.
To do so, we created the following question in the Cindicator app:
"The aggressively pro-US business approach of the Trump administration
could prompt the Federal Reserve to hike rates
more rapidly. Will the Fed raise the interest rate on Wednesday,
14 June?"
In this case, the Hybrid Intelligence deemed the probability of a rate hike
to be 66%, and in fact the Fed increased the rate at that meeting.
Regarding macroeconomic events, it is also possible to nd out whether a
macroeconomic indicator surpasses of analysts' expectations. For example, we
are going to make a trade based on the publication of the ADP Employment
Report, so we publish the following question in the Cindicator app:
"Last month ADP employment change was 172,000. The next

report will be published on Wednesday, 3 August. Will employment
rise by more than 165,000?"
21
In this question, Cindicator forecasters predicted a 67% probability of ADP
employment increasing by more than the consensus set. Indeed, that day, ADP
employment change to 179,000. Thus, Hybrid Intelligence provided a correct
forecast, and the market reacted in the right direction.
Let's adduce some more examples of similar questions related to macroeconomic
events:
"Will the jobless rate in May be greater than or equal to 4.4%?"
"Will China's second quarter GDP exceed estimates of 6.6%
according to a report to be published on July, 14?"
"Bank of Japan (BOJ) Governor Haruhiko Kuroda reiterated
that the central bank is prepared to step up stimulus if needed,
while noting again that so-called helicopter money is prohibited.
Will BOJ cut interest rates to below -0.2% on the 28 July

meeting?"
Find out more about macroeconomic questions on our Medium blog, where
we share the signals we receive and information on the trades completed based
on this data:
- Nonfarm Payrolls: +3.7% in two minutes;
- Durable Goods Orders: How to Trade with Crowd Indicator?
- SPY: +0.3% on Fed's decision.
6.1.2 Corporate events
Speculation about whether a published report will exceed its consensus forecast
is also used to predict some corporate events (e.g., quarterly reports). For this
purpose, we put the following question to Hybrid Intelligence:
"Best Buy Co., Inc. (BBY) is set to report its second-quarter
2016 results on Tuesday, 23 August before the market opens.
Wall Street is expecting earnings per share (EPS) of $0.43. Will
Best Buy report EPS above Wall Street Consensus?"
Cindicator predicted that Best Buy's report would be better than the analysts'

consensus with a 73% probability. Indeed, the EPS published turned out
to be higher than the consensus, and the company's shares spiked up by over
10% in one day.
Corporate events include not only report publication, but also dividend
payment, conference calls, and presentations. You can see examples of trades,
which can be made with the use of these forecasts, on our Medium blog:
Apple (AAPL) Conferences:
- Apple: +0.5% within a few hours;
- Apple: +5% in a few days.
Quarterly and annual reports:
- Guess? +13.2% in one trade;
- Apple: How to Trade with Crowd Indicator? +5.66% in one trade;
- Netix: +17.7% in less than 24 hours.
22
6.1.3 Political events
Political events analysed on our platform include the election of top ocials and

parties in various countries, resignations, important meetings, certain political
decisions (e.g., imposing or extending sanctions, or countries joining various
coalitions), and others.
For example, prior to the rst round of presidential elections in France, we
posted the following question in our application:
"Less than two weeks before the vote in the rst round, the farright
candidate Marine Le Pen was leading the polls in the 2017
French presidential election. The rst round will be held on 23
April 2017. Will Marine Le Pen reach the second round?"
Cindicator's Hybrid Intelligence generated an answer of 80%, and indeed,
Marine Le Pen passed into the second round. After that, the following question
was posted in the application:
"The rst round of the 2017 French presidential election was
held on 23 April 2017. As no candidate won a majority, a runo
election between the top two candidates, Emmanuel Macron
and Marine Le Pen, will be held on 7 May 2017. Will Le Pen

win the 2017 French presidential election?"
On 6 May, 2017, Hybrid Intelligence deemed the probability of Marine Le
Pen becoming president as 34%. On 7 May, Emmanuel Macron was elected
President of France. Thus, Cindicator's Hybrid Intelligence accurately predicted
the results of the presidential elections in France.
However, the answers received from Hybrid Intelligence, both correct and
incorrect predictions, which need to be interpreted in the right way, can also be
used as trading signals. For example, prior to the US presidential election, not
many people believed in Donald Trump's victory, and Cindicator's forecasters
estimated his chances of winning as 31%. However, US opinion polls showed
that Trump and Clinton were keeping pace with each other. History shows
that in such cases, when public opinion is so absolutely against facts (which
show equal chances of both events), black swans often appear. In expectation
of speculation around the election on 8 November, 2016, our analysts asked
a number of questions and then completed several successful trades based on
those questions. For more information, please see our blog on Medium:

- Volatility index: +3.14% in one trade;
- E-mini S&P 500 futures: +13% in less than 4 hours.
Thus, the signals received from Hybrid Intelligence can be inverted in some
cases to complete protable trades.
6.2 Binary price-related questions
In addition to questions related to events, binary questions are used when there
is a need to determine price movement. For example, we need to know whether
the price of an asset will rise to a certain level. By asking such a question, we
receive an indicator that shows the probability of the price achieving this level.
For instance, on 7 October, 2016, we wanted an accurate price forecast
regarding oil futures. Therefore, we posted the following question in the Cindicator
app:
"Oil climbed above $50 a barrel for the rst time since June as
declines in U.S. crude inventories and OPEC's pledge to reduce
supply lifted hopes the global glut may clear. Will WTI futures
surge above $51.7 before 20 October?"

We received the signal that the probability of such a price movement was
58%. On 10 October, the price rose to $51.6; our traders closed their position
23
with prot of 1.9%, without further price increase expectations.
A question regarding the cryptocurrency FoldingCoin, put to Hybrid Intelligence
on 15 June, 2017, is another example of a price binary event.
"The cryptocurrency FoldingCoin (FLDC/BTC) rose more
than 16% and settled at 0.00000834 at 4 PM ET at Poloniex
exchange on Thursday, June 15. Will FoldingCoin surge above
0.00001030 before 25 June?"
Hybrid Intelligence answered "yes". Indeed, on 21 June, FoldingCoin exceeded
0.00001030. On 22 June, its price reached a peak of 0.00001642, showing
almost 100% growth over the price at the time of the question being posted.
Binary signals can complement each other to create a comprehensive picture
of the current market situation. Thus, the next example contains a binary
question that was asked right after Netix published its quarterly report and

aimed at understanding the term of an increase in share price on the report:
"Shares of Netix, Inc. (NFLX) closed the week ending 21 October
25.65% higher at $127.50 after the company reported far
stronger-than-expected earnings per share on Monday, October
17. Will Netix stock drop below $120 before November 5?"
Hybrid Intelligence estimated such a probability at 34%. Indeed, until
November 5, the price was above $120 per share.
Here's another example of making a trade using a series of questions. In
May 2017, the market saw an adjustment of cryptocurrencies following a rapid
increase. Cindicator posted a series of questions regarding Litecoin, which
provided us with a signal to buy.
"The cryptocurrency Litecoin (LTC/USDT) settled on $30 at
7:15 AM ET at Poloniex exchange on Tuesday, 6 June. Will
LTC/USDT drop below $22.50 before 25 June?
"The cryptocurrency Litecoin (LTC/USDT) settled at $28.32
at 4 PM ET on Poloniex exchange on Thursday, 15 June. Will

LTC/USDT rise above $32 before 21 June?"
According to Hybrid Intelligence, the answer to the rst question was no,
and the answer to the second question was yes. Thus, our traders purchased
Litecoin with a stop below $22.50 using these signals. For more information,
see our post on Medium: Litecoin: +80% in three weeks.
Here are some more examples of binary price-related questions:
- Oil Futures: Swing Trades with Crowd Indicator. +1.7% in a few days;
- Oil Futures: +2.3% in two days.
Binary signals show high accuracy: the average accuracy of this type of
signal is 76% (the model is trained with over 300,000 forecasts).
Trading with the use of various binary signals as a main or auxiliary tool
can provide interesting nancial results. Thus, from July 2016 to February
2017, we increased our model portfolio by 123% by using only binary signals
and trading on traditional markets.
24
Cindicator's model portfolio return (result of binary question interpretation),

18/07/2016  28/02/2017
You can see more examples of binary questions in this Google spreadsheet.
6.3 Price-related questions
Along with probability questions, Cindicator introduced price questions in January
2017. The questions were tested beforehand with a group of forecasters
in a special challenge in October-December 2016, when the signals produced
excellent results.
For example, we needed to get a trade signal and entry/exit prices based on
minimum/maximum signals for Alphabet Inc. (GOOG) for Monday, 6 March,
2017. On Friday, 3 March, we created the following question in our mobile app:
"Shares of Alphabet, Inc. (GOOG) closed at $829.08 on Friday,
3 March. In your opinion, what will be the maximum and the
minimum price of GOOG on Monday, 6 March during market
hours?"
Usually, in such events, forecasters can provide their prediction of the minimum
and maximum price of nancial assets before the market opens (in our

example, before 9:20 AM ET, 6 March). Right after the question closes (deadline),
the articial intelligence system synthesises accurate forecasts using machine
learning algorithms based on the accumulated statistics predicted by
forecasters. Our system uses forward and back-tests of the history of synthesised
predictions to calculate entry and exit points for GOOG shares with the
optimal risk/reward ratio.
In this example, the signal was very accurate: the predicted minimum was
$822.43, and the actual minimum was $822.40 (an error of 0.0036%); the predicted
maximum was $828.81, and the actual maximum was $828.88 (an error
of 0.0084%). The net prot for GOOG, after all transaction costs, amounted to
0.31% in two trades in one day. The charts below demonstrate the accuracy of
the signals (all price levels are calculated by Hybrid Intelligence before market
opening):
25
Another example: entry signals for SPY (SPDR S&P 500 ETF), 09/03/2017
Find more examples of min/max price signals on our blog:

- Amazon: +2.5% in a few days;
- Facebook: +0.66% in a few hours;
- Dollar Tree: +1% in just 2 minutes;
- Visa: +1% in just 20 minutes;
- Apple: +0.15% in one hour.
Similar to other Cindicator signals, min/max signals are universal and can
be applied to any assets: cryptocurrencies, stocks, futures, currency pairs, etc.
Also, generating these signals is possible for various time frames  e.g. daily,
weekly, monthly, or quarterly. Specic questions related to the crypto-market,
as well as questions with long-term time frames, is linked to higher volatility.
Our regular weekly crypto-signals are an example of how our technology
deals with high volatility. Twice a week, we receive Hybrid Intelligence signals
on the top three cryptoassets in terms of capitalisation: Bitcoin, Ethereum,
and Ripple. For example, here is one of questions related to Bitcoin:
"[#WEEKLY]. The cryptocurrency Bitcoin settled at $2469 at
2:00 PM ET on Wednesday, 14 June. In your opinion, what

will be the maximum and minimum price of BTC/USDT in
over the next ve days (from 10:00 AM ET on Thursday, 15
June until 9:59 AM ET on Monday, 19 June)?"
The signals received with regard to Bitcoin question were quite accurate
and are clearly represented on the following chart:
Bitcoin (BTC/USDT), 15/06/2017 - 19/06/2017
Similarly, the example regarding Ripple looks as follows:
"[#WEEKLY]. The cryptocurrency Ripple settled at $0.2627 at
2:00 PM ET on Sunday, 11 June. In your opinion, what will
be the maximum and the minimum price of XRP/USDT over
the next 4 days (from 10:00 AM ET on Monday, 12 June until
9:59 AM ET on Thursday, 15 June)?"
26
These signals also turned out to be accurate, which can be clearly seen from
the chart below:
Ripple (XRP/USDT), 12/6/2017 - 15/06/2017

The min/max signals for price levels can be used to build numerous trading
strategies or to serve as an indicator of important support and resistance levels.
By using a simple counter-trend strategy (purchase at the min level and sell at
the max level), we managed to reach a return of 11.68% on traditional markets
from October 2016 to March 2017. From October 2016 to February 2017 trades
were not carried out on a daily basis, so there were only 148 trading days. Thus,
the annual protability of this strategy was 19.8%.
The min/max signals return (selling at the max level, purchasing at the
min level), 26/10/2016 - 06/06/2017
Examples of minimum-maximum price signals can be seen in this spreadsheet:
link.
Such a strategy yields good results for crypto-assets, particularly Bitcoins.
Weekly min/max signals described in detail above showed about 40% yield over
the last six months (80% p.a.) with Bitcoin. At the same time, these signals
are a countertrend, and the fact that they perform successfully in a trendy bull
market is a matter of healthy optimism. Also noteworthy is the fact that this

strategy works well on the short side tooshort selling brought almost twice as
much prot as longs (buying an asset), 22.31% vs. 13.96%, which is recognised
as a solid gain in a bull market.
27
The min/max signals return for Bitcoins (selling at the max level, purchased
at the min level), 13/01/2017 - 03/07/2017
6.4 Planned new data types, signals and indicators
Since 1 July 2017, a new type of price-related question has been addedquestions
pointing to a single price level. These may be questions related to the open
and close prices of a nancial asset, the price of an asset at a given moment,
associated with reaching a certain level, and so on. For example:
"Shares in NIKE, Inc. (NKE) closed at $59 on Friday, 30
June. In your opinion, what will be the close price of NIKE
stock on Monday, 3 July?"
Since May 2017, we have also been testing a new type of signal on a group
of forecasters who are consistently in the top 2% of the monthly rating. Users

are required to choose assets that they think should be bought or sold (short
selling), as well as assets for which transactions should not be made. Users
should also provide stop loss and take prot prices for those assets that they
choose to buy or sell. For example, the question is formulated as follows:
"What should we do with the cryptocurrency Ethereum
(ETH/USD) in the coming week: buy, sell or nothing? The
position will be opened at 9:30 AM ET on Monday, 22 May,
and will be closed at 4:00 PM ET on Friday, 26 May."
All forecasters provide their predictions for this asset. If they choose an action
(buy or sell), they also need to set stop-loss and take-prot orders for their
position. In this way each person can 'manage' their own investment position,
choosing the expected direction and risk/prot parameters. For example, one
of the users answers that Ethereum should be sold with a stop-loss of 199.9
and take-prot of 145.1; the second user chooses an option of buying (with
a stop-loss of 120 and take-prot of 290); a third user indicates that nothing
needs to be done, and so on.

Then it's the articial intelligence algorithm's turn: it aggregates the forecasters'
predictions into a single one, determining which action should be taken
and with what parameters. In our example, Hybrid Intelligence decided that
this week Ethereum should be bought with a take-prot of 196.62 and stoploss
of 129.78, on the condition that the position opens at the opening price at
9:30 AM ET on Monday. If neither the take-prot nor stop-loss is achieved,
then the position will be closed on Friday. Indeed, Ethereum price increased,
and this signal brought the model portfolio a prot of more than 32% over the
course of ve days.
For one month the prot on such signals was +12% in the stock market and
+86% in the market of crypto-currencies (only three crypto-assets were used
in trading: Bitcoin, Ethereum, and Ripple). We consider it an excellent result
and plan to add such questions to the application in the near future.
The development of signals in which the task of Hybrid Intelligence is to
rank assets on a certain basis is being actively pursued. For example, the companies
conducting an ICO in the nearest month need to be arranged according

28
to the degree of their probable success or trustworthiness. Each forecaster gets
100 points and the list of planned ICOs in the current month, as shown in the
example below:
- AEternity
- Bancor
- Civic
- Cofound.it
- Monaco
- OmiseGO
- SONM
- Status
The forecaster should then rank the proposed ICO based on the degree of
expected success (expressed, for example, in the growth of the price of a token
over a certain period) or by personal trust therein. To do this, the forecaster
distributes the 100 points among them, giving more points to companies expected

to achieve greater success with the ICO.
The articial intelligence system then aggregates the scores from all forecasters
and determines which ICOs will be the most successful.
Tests for the stock market held in November 2016 (where forecasters were
required to rank the shares of dozens of companies based on expected growth
in their share price for the coming week) showed the consistency of these signals
as a standalone indicator and as one of the support indicators in making
investment decisions.
Another promising area of our research are signals in which forecasters need
to specify a date in answering the question. For example, we are interested in
this type of event:
"When do you think any G20 country will issue a national cryptocurrency?"
Or this event:
"On what day will Ethereum break $1000?"
A forecaster will name the date, and articial intelligence, in its turn, will
aggregate all predictions of forecasters into a single signal.

6.5 Planned new analytical products
The experiments we conducted and the results achieved by our colleagues studying
various aspects of the operation of collective intelligence will enable us to
create the following set of analytical products, which will be integrated into
the overall capital management infrastructure:
- Assessing the power and inuence of news on markets. Forecasters are
oered N potential news hooks or events that will occur in the future and
required to assess the direction of motion (whether this may lead to an
increase or a drop in the price of an asset), as well as the impact of such
motion;
- Uniting forecasters in centralised groups (depending on the predictions'
accuracy in a particular cluster) and linking these groups to a decentralised
structure (forecasters in a group collectively make predictions,
which are further aggregated by articial intelligence between dierent
groups to create dierent data sets and trading signals);
29

- Access to analytics from superforecasters (signals exclusively from the
top 2% ranked forecasters; channels for direct interaction with superforecasters
for any category of interest, e.g., long-term Bitcoin analytics);
- Analysis of the existing investment portfolio of a particular trader by
collective intelligence;
- A thermal geographic map of the market showing anticipated price increase/decrease
for a certain asset. The map is compiled on the basis
of predictions of forecasters from dierent regions. For example, US
users taken together expect a 40% increase in Bitcoin over the next quarter,
citizens of China expect 80%, and citizens of Germany only 5%.
Thus, the map will explicitly show the distribution of expectations for
the growth/decline of various assets.
6.6 Public experiment with Moscow Exchange
In January 2017, we launched a public experiment with Moscow Exchange
(MOEX), one of Europe's biggest exchanges. Moscow Exchange is also one of
the ten largest exchange platforms for derivatives trading globally (according

to the World Federation of Exchanges monthly report statistics).
In this experiment, we completed trades based on the synergy of articial
intelligence and the collective intelligence of people with varied professional experience,
united by a common goal. In addition, every participant had personal
motivation.
To carry out this experiment, we collected a new sample of 925 unrelated
people: 40% of the participants had never made trades on the exchange before;
60% had various investment experiences.
We received min/max levels (25,000 forecasts) from our forecasters regarding
four nancial assets (USD/RUB futures, Brent oil futures, silver futures,
and gold futures) for three weeks on a daily basis, excluding weekends and
holidays. After receiving forecasts, the robot aggregated them into a signal
(including the entry level, stop loss, and take prot) and simulated the trades.
As the experiment was public, we posted all entry levels to our Telegram
public channel on a daily basis prior to market opening.
Based on the forecasts, Cindicator's robot modelled 57 trades, 36 of which

were protable, over the course of the experiment. The completed trades
showed a 3.0% return in 29 days, which is equal to 26% p.a.
The result of this pilot proved the eciency of the Hybrid Intelligence
ecosystem even with a non-trained (on the computer learning models used
by us) sample of forecasters.
Total gain, public experiment with Moscow Exchange from 19/01/2017 to
14/03/2017:
30
Entry/exit signals for USD/RUB exchange rate futures, 01/03/2017
Entry/exit signals for Brent oil futures, 27/02/2017
Entry/exit signals for Gold futures, 10/03/2017
Article highlighting the results of a public experiment with the Moscow Exchange
(original post on the website of the Moscow Exchange).
Trades made during public experiment.
7 Team and stages of development
7.1 Team

The Cindicator team has been created by a synergy of like-minded people with
a variety of expertise. About 85% of the team members are graduates of top
STEM universities.
In 2014, Mike Brusov had the idea of creating an application for users where
they can make various nancial, political, and sports forecasts while competing
with one another. Studying collective intelligence and the data generated
by numerous diverse individuals was one of his professional activities in 2010.
From 2010 to 2013, Mike, as one of the Wobot startup founders (an automated
service for monitoring and analysing data generated by users in social media)
studied the correlations between media coverage of numerous Internet posts
that contained diering sentiments and realities.
31
In 2015, Mike collected the required initial investment and invited his future
partners, Yury Lobyntsev and Artem Baranov, to join him in exploring
the potential of collective intelligence.
Yury Lobyntsev has been working with computer systems and programming

large-scale software systems since early childhood. At the same time he
has been exploring the phenomena of intelligence, mind, and consciousness.
In 2011, Yury founded his rst tech business, the Oumobile studio, which is
engaged in tech development for mobile startups.
In 2012, Artem Baranov established a digital product development company,
which emerged from popular St. Petersburg local app Most Have. He is
distinguished by his sense of style, laconic manner, and ability to apply new
technologies in everyday life. Steady progress of the business required growth
and strengthening of the companys position on the digital product development
market - this was the reason for merging with Yury's team in 2014.
In 2014, Yury and Artem jointly established the Octabrain neurolaboratory,
which undertook investment and R&D for startups and innovation projects.
The company was working in the eld of neurointerfaces (braincomputer interfaces),
and neural networks. Through Octabrain, they organised a team of
neuroscientists and digital product developers to explore and invent intelligent
systems for human-machine interaction.

Mike, Yury, and Artem then united and began studying collective intelligence
systems. After the release of the rst version of the collective forecasting
platform in December 2015, it was conrmed that among a random group of
individuals there were 2% who very accurately predicted the answers to certain
types of questions. Researchers Philip Tetlock and Den Gardner called these
individuals superforecasters in their studies. It became clear that the wisdom
of crowds' accuracy would be signicantly enhanced by a statistical analysis of
these forecasters accuracy.
Therefore, in 2016 the Cindicator team began growing its expertise in data
science and machine learning. Data scientist Alexander Frolov was the rst
one who joined the team. After the arrival of the mathematician, teacher of
mathematical methods, and backend developer, the technical team began to
strengthen its mathematical and data analysis expertise in the domains of data
processing and articial intelligence modeling.
Following the decision to focus on nances and investments while building
the ecosystem, Kate and Nodari joined the team: highly experienced in

managing large positions, trading global stock, futures, foreign exchange, and
cryptocurrency markets, these traders have worked in nance for over ten years.
Their comprehensive expertise allows Cindicator to seek new ways to apply Hybrid
Intelligence in various spheres of nancial markets.
7.2 Current progress of the company
December 2015 saw the global release of the rst iOS-based version of the platform.
This was a step forward towards the creation of a collective intelligence
system of the required size.
In January 2016, the team was invited to several startup accelerators and
chose Starta Accelerator due to its location in the heart of the global nancial
system - New York. Once the acceleration program was successfully completed,
Cindicator raised $300,000 during the pre-seed round of venture investment.
From June to November 2016, the team worked on the creation of the rst
set of machine learning models, the further improvement of the collective intelligence
system (currently, there are over 8,000 forecasters on the platform),
and the creation and forward testing of various trading strategies on the stock

and foreign exchange markets. Intermediate results conrmed a number of hy-
32
potheses and provided interesting results.
In January 2017, an API for trading signals was launched, enabling the
launch of test integrations with 11 hedge funds and three banks in the following
four months. At that time, platforms were supplemented with the rst
crypto-assets, followed by internal experiments and related forward tests.
From November 2016 to March 2017, Cindicator took part in the rst batch
of the Moscow Exchange ntech incubator, where it was ranked as the topperforming
startup. The company was granted $120,000 for technology development
from Microsoft and became a member of the Microsoft BizSpark
startup support program.
In April-May 2017, the company attracted $200,000 from a number of ntech
investors in a seed venture round.
8 Legal considerations
8.1 Legal

We are have approached the Cindicator token sale in a comprehensive and responsible
manner. Given the uncertain status of cryptocurrency and digital
tokens in various jurisdictions, we spent a signicant amount of time and resources
to analyze the legal status of Cindicator business model and the CND
tokens in jurisdictions where we plan to operate. In the United States, we
worked closely with Velton Zegelman PC, a Silicon Valley law rm actively
representing blockchain and cryptocurrency clients. In Gibraltar, the jurisdiction
of Cindicator Ltd (Gibraltar) (in formation as of August 15, 2017) we plan
to work with ISOLAS, a leading and oldest law rm in Gibraltar.
Due to the uncertain state of regulation on a global scale, we cannot guarantee
the legality of Cindicator hybrid intelligence platform or ability to structure
and license a future investment fund based on our platform in any given jurisdiction.
However, we strive to be a responsive and compliant company should
we face any regulatory inquiry.
8.2 Legal status of CND tokens
CND tokens are functional utility tokens designed for the Cindicator hybrid

intelligence platform. CND tokens are not securities. Once you purchase CND
tokens, they cannot be refunded. We do not recommend buying CND tokens for
speculative investment purposes. You should buy CND tokens to participate
in the Cindicator hybrid intelligence platform. CND tokens are not equated
with participation in Vote, Inc. and/or Cindicator Ltd (Gibraltar) and CND
token holders have no equity, governance, or any other rights in either company.
CND tokens are sold as a digital asset, similar to downloadable software, digital
music, and alike. We do not recommend purchasing CND tokens unless you
have prior experience with cryptographic tokens and blockchain-based software.
8.3 Legal status of crowdsourced forecasting platforms
There is no unied regulatory framework applicable to crowdsourced forecasting
platforms. These products and services are regulated in some jurisdictions
based on existing gaming and/or nancial services regulatory frameworks, while
they are left unregulated in others. Before targeting a particular jurisdiction,
we will conduct legal due diligence analysis of applicable regulations in such
jurisdiction. Depending on the regulatory burden and steps involved, we will

then either take the necessary steps to obtain any required licenses and/or permits
in such jurisdiction or withhold from operating in such jurisdiction.
33
For the convenience of our users, Cindicator White Paper, website and other
related documents are available in a number of languages. In the event there
is any conict between the English language version and a foreign language
version, the English language version shall govern.
9 Conclusion
Cindicators ultimate goal is to set up a decentralised intellectual technology
that eectively implements the potential of Hybrid Intelligence for the benet
of all participants of the ecosystem. In the future the technology strives to
be fully automated: the only resource necessary for it to function will be the
mental investment by the analysts.
Hybrid Intelligence anticipates being used not only in nancial and economic
markets, but also in art, politics, sports, business, technologies, and
science in the future.

Cindicators token sale is an excellent opportunity to join the development
of a symbiotic relationship between the minds of people and machines.
10 Risk factors and disclaimers
THIS DOCUMENT DOES NOT CONSTITUTE AN OFFER TO SELL, AN
INVITATION TO INDUCE AN OFFER, OR A SOLICITATION OF AN OFFER
TO ACQUIRE SECURITIES. THIS DOCUMENT IS PROVIDED FOR
INFORMATIONAL PURPOSES ONLY AND DOES NOT CONSTITUTE INVESTMENT
ADVICE.
THE SALE OF CND TOKENS CONSTITUTES THE SALE OF A LEGAL
SOFTWARE PRODUCT UNDER GIBRALTAR LAW. THIS PRODUCT
SALE IS CONDUCTED BY CINDICATOR LTD* (GIBRALTAR), A
GIBRALTAR PRIVATE LIMITED COMPANY, OPERATING UNDER GIBRALTAR
LAW. IT IS THE RESPONSIBILITY OF EACH POTENTIAL
PURCHASER OF CND TOKENS TO DETERMINE IF THE PURCHASER
CAN LEGALLY PURCHASE CND TOKENS IN THE PURCHASERS JURISDICTION
AND WHETHER THE PURCHASER CAN THEN RESELL

THE CND TOKENS TO ANOTHER PURCHASER IN ANY GIVEN JURISDICTION.
ALL POTENTIAL RISKS CAN BE ASSESSED HERE.
OUR WHITE PAPER MAY CONTAIN 'FORWARD LOOKING STATEMENTS'
- THAT IS, STATEMENTS RELATED TO FUTURE, NOT PAST,
EVENTS. IN THIS CONTEXT, FORWARD-LOOKING STATEMENTS OFTEN
ADDRESS OUR EXPECTED FUTURE BUSINESS AND FINANCIAL
PERFORMANCE, THE PERFORMANCE, AND ACCURACY OF CINDICATOR
HYBRID INTELLIGENCE PLATFORM, AND OFTEN CONTAIN
WORDS SUCH AS 'EXPECT', 'ANTICIPATE', 'INTEND', 'PLAN', 'BELIEVE',
'SEEK', 'SEE', 'WILL', 'WOULD', 'ESTIMATE', 'FORECAST' OR
'TARGET'. SUCH FORWARD LOOKING STATEMENTS BY THEIR NATURE
ADDRESS MATTERS THAT ARE, TO DIFFERENT DEGREES,
UNCERTAIN. WE CANNOT GUARANTEE THAT ANY FORWARD LOOKING
STATEMENTS, BACKTESTS OR EXPERIMENTS MADE BY US OR
EXPECTED RESULTS OF OPERATION OF CINDICATOR HYBRID INTELLIGENCE
PLATFORM WILL CORRELATE WITH THE ACTUAL FUTURE

FACTS OR RESULTS.
FOR THE CONVENIENCE OF OUR USERS, CINDICATOR WHITE
PAPER, WEBSITE AND OTHER RELATED DOCUMENTS ARE AVAILABLE
IN A NUMBER OF LANGUAGES. IN THE EVENT THERE IS ANY
CONFLICT BETWEEN THE ENGLISH LANGUAGE VERSION AND A
FOREIGN LANGUAGE VERSION, THE ENGLISH LANGUAGE VERSION
34
SHALL GOVERN.
*LEGAL ENTITY IS CURRENTLY IN FORMATION IN GIBRALTAR
35
References
[1] The Wisdom of Crowds (The Wisdom of Crowds: Why the Many Are
Smarter Than the Few and How Collective Wisdom Shapes Business,
Economies, Societies and Nations - James Surowiecki, 2004).
https://en.wikipedia.org/wiki/The_Wisdom_of_Crowds
[2] The Good Judgment Project (Philip E. Tetlock, Barbara Mellers, Don

Moore). https://en.wikipedia.org/wiki/The_Good_Judgment_Project
[3] Intelligence Advanced Research Projects Activity.
https://en.wikipedia.org/wiki/Intelligence_Advanced_Research...
[4] Superforecasting: The Art and Science of Prediction (Philip E. Tetlock,
2015). https://en.wikipedia.org/wiki/Superforecasting
[5] Iowa Electronic Markets. https://en.wikipedia.org/wiki/Iowa_Electronic...
[6] Delphi method. https://en.wikipedia.org/wiki/Delphi_method
[7] Reference class forecasting. https://en.wikipedia.org/wiki/Reference...
[8] Consensus forecast. https://en.wikipedia.org/wiki/Consensus_forecast
[9] Shubharthi Dey, Yash Kumar, Snehanshu Saha, Suryoday
Basak. Forecasting to Classication: Predicting the direction
of stock market price using Xtreme Gradient Boosting.
https://www.rsearchgate.net/publication/309492895_Forecasting_to...
[10] Shunrong Shen, Haomiao Jiang, Tongda Zhang. Stock
Market Forecast-ing Using Machine Learning Algorithms.
http://cs229.stanford.edu/proj2012/ShenJiangZhang-StockMarket...

[11] Ina Khandelwal, RatnadipAdhikari, GhanshyamVerma.
Time Series Forecasting Using Hybrid ARIMA
and ANN Models Based on DWT De-composition.
http://www.sciencedirect.com/science/article/pii/S1877050915006766
[12] Erhan Bayraktar, H. Vincent Poor, K. Ronnie Sircar. Estimating the
Fractal Dimension of the S&P 500 Index using Wavelet Analysis.
https://www.princeton.edu/ sircar/Public/ARTICLES/bps.pdf
[13] Takeshi Inagaki. Critical Ising Model and Financial Market.
https://arxiv.org/abs/cond-mat/0402511
[14] Robert Nau. ARIMA models for time series forecasting.
https://people.duke.edu/ rnau/411arim.htm#pdq
[15] Paulo Rotela Junior, Fernando Luiz Riera Salomon, Edson
de Oliveira Pamplona. ARIMA: An Applied Time
Series Forecasting Model for the Bovespa Stock Index.
https://le.scirp.org/pdf/AM_2014120514194065.pdf
36