Early Community Draft Version 0.3 - Subject to Change

Bloom ProtocolDecentralized credit scoring powered by Ethereum and IPFS

Bloom is a protocol for assessing credit risk through federatedattestation-based identity verification and the creation of anetwork of peer-to-peer and organizational creditworthinessvouching (“credit staking”)

Jesse Leimgruber, Alain Meier, John Backus

Last Updated January 27, 2018

Table of Contents

1. Abstract 1

2. Bloom Protocol Overview 3

3. BloomID (Identity Attestation and Credit Vouching) 3

Organizational Identity Attestation . . . . . . . . . . . . . . . . . . . . . . . 4

Peer-to-Peer Attestation and Vouching . . . . . . . . . . . . . . . . . . . . . 5

4. BloomIQ (Credit Registry) 7

Payment Data Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Tamper-Proofing Reported Data . . . . . . . . . . . . . . . . . . . . . . . . 8

Fair Credit Reporting by Default . . . . . . . . . . . . . . . . . . . . . . . . 8

Risk Evaluation and Reliability Scores . . . . . . . . . . . . . . . . . . . . . 9

6. BloomScore (Credit Scoring) 9

Scoring Phase 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

Calculating Reliability Scores . . . . . . . . . . . . . . . . . . . . . . . . . . 10

Scoring Phase 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

Scoring Phase 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

7. The Loan Risk Assessment Lifecycle 11

8. Bloom Token (BLT) 11

9. Roadmap 13

10. BloomCard 13

11. Team 14

Founding Team . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

Advisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1. Abstract

Background

In 2015, the US Congress declared credit scoring to be a monopoly controlled by justone organization, FICO[1]. FICO provides credit scoring for more than 90% of topUS lenders[2]. FICO’s credit scoring system leaves over 26 million Americans “creditinvisible” and an additional 19 million unscorable[3].

Globally, the situation is even worse. 38% of the world’s population does not havea bank account[4]. 3 billion people are unable to obtain a credit card and 91%of residents in developing nations experience difficulty receiving debt financing fromtraditional financial institutions. Traditional credit bureaus require borrowers to takeon debt before obtaining a credit score, leaving millions of potentially creditworthyindividuals unscorable by the current credit system.

Credit scoring is similarly siloed around the world, further exacerbating these issues.Credit scoring providers can not operate globally, meaning that when a borrowermoves to a new country, they must rebuild their credit scores from scratch as theirscore does not follow them. Since identity verification is also centralized, applying fora loan requires users to expose all of their personal information, putting individualsat increased risk of experiencing identity theft. Credit losses due to identity theftexceed $21 billion each year.

Overview

In this whitepaper, we introduce a global, decentralized credit protocol, Bloom.Bloom addresses these existing limitations in lending by moving credit scoring andrisk assessment to the blockchain.

Bloom is a standardized, programmable ecosystem to facilitate on-demand, secure,and global access to credit services. Bloom presents a novel approach to credit riskassessment allowing both traditional fiat lenders and digital asset lenders to issuecompliant loans on the blockchain while increasing competition to lower fees andimprove borrower experience at every layer of the credit issuance process.

The Bloom protocol presents solutions to the following problems:

1. Cross-Border Credit Scoring: Credit histories are not portable across coun-tries, forcing individuals to re-establish their credit track records from scratchwhen they relocate.

2. Backward-Looking Creditworthiness Assessment: Credit systems rely onhistorical debt repayment information and therefore cannot easily accommodateusers who are new to credit. This is especially prevalent among minorities, theunderbanked, and the youth[5].

1

3. Lenders Have Limited Ability to Expand and Offer Loans Globally:Borrowers in markets with less developed financial and regulatory infrastructurestruggle to access credit as lenders have limited identity and scoring data to basecredit decisions.

4. High Risk of Identity Theft: Borrowers must expose all of their personalinformation when applying for a loan - the same info an attacker can use toopen new lines of credit.

5. Uncompetitive Credit Scoring Ecosystem: Credit data is centralized. Inmost markets, a single provider scores credit, resulting in an uncompetitiveecosystem for evaluating credit risk. FICO was checked on 90% of all U.S.Loans[2].

Protocol Components

There are three main systems which comprise the Bloom protocol:

1. BloomID (Identity Attestation): BloomID creates a global secure identity,allowing lenders to offer compliant loans globally, without forcing borrowers toexpose personal information.

2. BloomIQ (Credit Registry): BloomIQ is a system for reporting and trackingcurrent and historical debt obligations that are tied to a user’s BloomID.

3. BloomScore (Credit Scoring): The BloomScore is a metric of consumers’creditworthiness. This decentralized score is similar to FICO or VantageScorescore, but with updated models.

The Bloom protocol improves the current credit ecosystem by creating a globallyportable and inclusive credit profile, reducing the need for traditional banking in-frastructure and opaque, proprietary credit scores. This means both traditional fiatlenders and digital asset lenders will be able to also securely serve the 3 billion peoplewho currently cannot obtain a bank account or credit score.

Bloom decentralizes the credit industry while lowering rates and increasing security.Bloom makes it easy for lenders to transition to the blockchain by offering a new,compliant way for them to access new markets.

2

2. Bloom Protocol Overview

The Bloom protocol facilitates the broadening andefficient operation of the credit market by allowingboth fiat and digital asset lenders to extend creditto individuals and institutions operating in marketswith underdeveloped or immature credit infrastruc-ture, national identities or banking systems, with-out taking on additional risk.

The Bloom protocol provides solutions to enableany lender authorized by a borrower to safely andsecurely issue credit to that borrower. There arethree components of the Bloom protocol:

BloomID, Attestation-Based Identity andCreditworthiness

BloomID lets users establish a global, federatedidentity with independent third parties who pub-licly vouch for their identity information, legalstatus and creditworthiness. These third partiescan be friends, family or peers who vouch for auser’s identity and/or creditworthiness (“peer-to-peer staking”) or organizations who earn revenueby evaluating a user’s credentials (“organizationalstaking”).

Organizational stakers can either inherit trust fromtheir existing reputation (such as existing creditbureaus and identity companies who publicly an-nounce which attestation contracts they control) orby establishing a track record of successful identityattestations on the network.

BloomScore, A Decentralized Credit Scoreon the Blockchain

BloomScore is a dynamic indicator of an individ-ual’s likelihood to pay debts that adapts to the ma-turity of a user’s credit history (or lack thereof).By splitting a user’s credit scoring mechanism intothree phases that each take into account differentdata points with varying weights, BloomScore canproduce a score that is conducive to building creditfrom the ground up while helping creditors differ-entiate the credit risk of consumers in markets andcommunities with sparse data.

BloomScore initially relies on peer-to-peer creditstakes from BloomID as a means of bootstrappingtrust and eventually transitions to primarily usingthe user’s own spending habits and credit activ-ity as a proxy for creditworthiness. Organizationalstakers with an established presence can vouch forusers with strong credit under the current systemallowing them to transition their existing historyand reputation to the Bloom network.

BloomIQ, A Registry for Reporting andTracking Historical Credit

BloomIQ is a system for reporting and tracking cur-rent and historical debt obligations that are tied toa user’s BloomID. BloomIQ’s tracking mechanismputs the user in control, requiring each instanceof payment data release to a 3rd party to be au-thorized by the user. One of the primary goals ofBloomIQ is to allow a user to import existing credithistory to this decentralized system, reducing theneed for credit-established users to build up theircredit quickly.

3. BloomID (Identity Attesta-tion and Credit Vouching)

The foundation of a decentralized credit system is asecurely established and verified identity. In orderto prevent against common network attacks (suchas Sybil attacks), each participant’s identity must

3

be adequately established with a high cost of at-tempting to create new, false identities that appearauthentic.

BloomID is the Bloom protocol’s method of bothestablishing reliable identity as well as forming thebasis of creditworthiness for users who are newlyentering the Bloom network. BloomID allows or-ganizations who store information about individualidentities to attest to the identity of a Bloom userand mark that information on the blockchain forfuture re-use.

In addition, a user’s friends, family and peers canhelp an individual bootstrap creditworthiness byvouching for their ability to act responsibly withcredit. During this credit vouching process, a user’speers can also attest to various identifying traitssuch an individual’s name or date of birth to helpsecure the network and provide further proof ofidentity for the user.

Organizational Identity Attestation

Private companies like credit bureaus and govern-ments currently hold the majority of identity datathat exists in the world. These organizations play acritical role in helping onboard users onto the Bloo-mID system by attesting to the authenticity of user-submitted data.

In a loan contract, the lenders detail the identityattesters whom they trust and would require to at-test to the potential customer so that they can ful-fill the loan within their risk parameters. Theseexternal attesters can then agree to verify the iden-tity information and provide their public keys inresponse along with a description of the data theyneed in order to complete their attestation, such as“name”, “date of birth”, and “address”. The userthen attaches their identity information encrypted

for each identity attester respectively using theirpublic keys. The attesters evaluate the informa-tion returned to them by the user and publish onthe blockchain whether it has satisfied their require-ments.

Reusable Identity Verification

By publishing all historical identity attestations onthe blockchain, organizations can help take part inbuilding a reusable identity that builds up trustover time rather than having to be re-evaluatedfor every transaction with a new lender. This cannot only save money across the network of lenders,but it can also help significantly reduce on-boardingtime by reducing duplicate work by anti-fraud andcompliance teams across lending organizations.

Example Organizational Identity Attesta-tions

There are many kinds of identity attesters thatcan be supported by BloomID. Below is a non-exhaustive list of potential attestation types:

1. Electronic ID Verification: Verification ofan identity data by cross-checking supplied in-formation with a multitude of public records,private records and governments from aroundthe world.

2. Documentary Verification: Verification ofan identity document like a passport or adriver’s license and whether the image of theperson on the document matches the user sub-mitting the scan of the document.

3. Social Verification: Verifying the identityinformation of users via social networks likeFacebook and analyzing their friend graph tohelp reduce fraud.

4. Sanction Screening: Ensuring that a useris not on one of the many global sanctionprograms operated by various governmentsaround the world.

5. Politically Exposed Persons: Ensuringthat a user is not considered to be a politi-cally exposed person (someone with a promi-

4

nent political function who is at high risk ofpotential bribery or corruption involvement).

Peer-to-Peer Attestation and Vouch-ing

In the Bloom protocol, “peer-to-peer staking” isa mechanism for representing real-world relation-ships between individuals with the goal of estab-lishing both an indicator of creditworthiness andauthenticity of identity. Evidence suggests thatan individual’s creditworthiness can be reliablydetermined by the people who would vouch fortheir creditworthiness[6]. This concept of vouch-ing would not be a specific statement about a creditevent such as “Bob vouches that Alice is likely to re-pay a $10,000 loan”, but rather a general statementthat Bob trusts Alice’s judgement to not apply formore credit than she can afford.

Conceptually, this is not dissimilar to Google’sPageRank algorithm, at the core of which is theassumption that if a webpage has a high count ofquality inbound links, then the webpage must beimportant.

Enabling users to vouch for (“stake”) other peoplethey personally know, and whom they expect to befinancially responsible has several benefits:

1. Facilitates access to credit for new users2. Increases permanence in credit scoring and

ensuring appropriate weighting of long-termand short-term spending and payment habits,e.g. by avoiding users creating new accountsto reset their score

3. Enhances resilience to fraud through theBloom network

Revealing Creditworthiness Through Rela-tionships

By establishing bilateral stakes with trusted col-leagues, friends and family, users reveal the typeof financial network they are a part of. If a userdoes not have a rich credit history of their own,

the financial history of their peers can be used as aheuristic and indicator for what patterns their ownrepayment behavior can reasonably be expected tofollow[7]. FICO, by comparison, considers indi-viduals who lack a credit history as “credit invis-ible”. The purely retrospective approach by defini-tion renders FICO (inter alia) unable to evaluate anew customer.

For example, if Alice stakes four of her peers whohave all paid off their student loans and pay theirrent on time, then a lender can take more of a riskon Alice because she is likely to behave similarly toher peers. Likewise, if Frank stakes several friendswhen he joins the Bloom network and these friendsend up making late payments and / or defaultingon loans, then a lender can mitigate their risk indealing with Frank by requiring a higher interestrate or requesting collateral.

In staking acquaintances who ultimately fail tomeet obligations towards their creditors, Frank hascompromised his own creditworthiness. Similarly,Frank’s acquaintances have imposed a cost on himby failing to adequately meet their financial obli-gations. The network effect created by this systemextends beyond enhancing accountability betweencreditors and debtors, but adds an additional el-ement of accountability amongst real-world socialgroups.

If Frank realizes that most of his peers are unlikelyto behave financially responsibly, he might be moreconservative with the number of peer-to-peer stakeshe establishes. Staking a small number of peersmight be done to deliberately hide the fact that theuser does not know many people who are financiallyresponsible, but this reduced participation in the

5

peer-to-peer staking network will be factored intothe assessment of Frank’s own creditworthiness.

These varying behaviors that produce a differentnumber of stakes can be used as a heuristic for cred-itworthiness. A user who is part of a network offinancially responsible participants should be ableto quickly accrue bilateral stakes. Each of the usersthey stake is also likely to have more stakes of theirown that have been acquired quickly. The expecta-tion is that members of a riskier social grouping aremore likely to add stakes slowly and be in a moresparsely connected network.

Identity Attestation and Furthering Perma-nence of Metrics

If a user has a low BloomScore and has difficulty ob-taining credit, they will likely be tempted to aban-don their BloomID and create a new one. Bilateralstaking imposes a cost on such behavior, making itmore detrimental for a user to abandon their Bloo-mID than to simply accept their current Bloom-Score.

When Alice and Bob stake each other, the Bloomapp can prompt each user to confirm the otheruser’s birthday. Then, each user’s stake can includea hash composed of a secret generated by the stak-ing user and the birthday of the user they are stak-ing. This creates an identifier unique to the stakerand stakee that should remain the same if the re-cipient signs up for a new account with the samebirthday. In turn, we can enforce that the user’sbirthday be the same across accounts by requiringthat the signup birthday be provided as part of anattestation for a loan.

A user could try to stake an entirely different setof people with their new account, but this wouldrequire significantly more work and the user’s in-centive for their first account would have been toestablish as many quality stakes as possible to im-

prove their own creditworthiness. This leaves theuser with the options of staking users they stakedwith their first account, staking peers they did notwant to stake with their first account, or staking sig-nificantly fewer people with their new account. Ifthe user stakes peers they chose not to stake withtheir first account, or stakes significantly fewer peo-ple, their new account will have a harder time get-ting access to credit as well. If the user choosesto stake even a subset of the same users that theystaked with their first account, the network will beable to identify that several peer-to-peer stakes areshowing up with the same attached secrets and flagthe account as a duplicate.

This system acts as a valuable heuristic to ensurethat fewer users will increase their perceived credit-worthiness by abandoning their BloomID and mov-ing to a new one, thereby enhancing the perma-nence of the BloomScore.

Securing the Bloom Network

An attacker could create hundreds of fake Bloo-mIDs and have these fake accounts all stake eachother. The attacker could then setup her own fakeloan organization and have the fake accounts pre-tend to take out and pay off loans. This processcould effectively produce BloomIDs that appear au-thentic with good financial history that the attackcould then use to defraud real loan originators.

The Bloom network will bootstrap the network bymarking a small number of users and organizationsas “trusted” network participants. These partici-pants will be manually vetted by the Bloom team.Loans, attestations, and risk assessments involv-ing trusted participants during the bootstrappingphase will mark certain participants as authenticBloom users. Trusted users will be instructed toonly stake people they know and are willing tovouch for, so they are unlikely to stake the fakeusers created by our attacker. Likewise, unless theattacker is willing to create and maintain a real loancompany, the real users are unlikely to interact withthe fake loan companies.

6

Networks of real users should contain marked par-ticipants from the bootstrapping phase. The at-tacker’s network will seem unusually disconnectedfrom the rest of the network. If the attacker is suc-cessfully engaged in a loan scam with one of her fakeusers, it will taint the scores of many users in theattacker’s network. An unusually disconnected net-work of users with financially responsible users thatsuddenly start scamming loan companies will beidentifiable quickly before it can cause a meaning-ful amount of damage to the network. Peer stakingdoesn’t solve fraud, but it dramatically increasesthe cost of abuse within the Bloom network.

4. BloomIQ (Credit Registry)

BloomIQ is a system for reporting and tracking cur-rent and historical debt obligations that are tied toa user’s BloomID. BloomIQ is designed to bring thewealth of pre-existing and comprehensive credit his-tory to the blockchain while maintaining privacy forthe user by introducing a user approval-based sys-tem of information dissemination, offering a markedimprovement over current systems. Data aboutan individual’s ability to pay past debts remainsan important part of determining credit risk, andBloomIQ enables this functionality to be decentral-ized and reusable.

For example, when issuing a loan to Alice, a lendercan see Alice’s:

1. Reliability score (a metric gauging a user’s in-dividual credit repayment history success)

2. Peer score (a metric to determine the averagereliability score of the peers of the user)

3. Number of loans taken out in the past on theBloom network

4. Past identity attestations performed

Unless Alice has taken out a loan from this com-pany before, they cannot see:

1. Alice’s past loan information (total loaned,payment amounts, etc)

2. Her peer’s transaction history3. Identifying information about Alice (name,

address, etc)

If Alice has enough past loans and still has a veryhigh BloomScore then the loan company could optto issue the loan without further checks. If the loanprovider wants more information though, they canset a requirement on the loan that a risk assessmentof their choosing (“RiskCo”) needs to stake the loanin order for the contract to release funds to Alice.

When RiskCo’s stake is added as a requirement tothe loan, RiskCo can

1. Ask for access to Alice’s existing payment his-tory

2. Add a requirement to the contract for a dataprovider (“DataCo”) to provide payment datathat isn’t available on the blockchain (for ex-ample, utility bill payments)

If DataCo needs information about Alice’s identityto lookup data then it would add a PII requirementstating that it needs Alice to share certain identi-fying information so it can perform a lookup via itsalternative data sources.

Alice has final say over all requests. She can choosenot to provide PII or payment information if shedoesn’t want to disclose it and it will simply ter-minate the contract without cost to any parties in-volved. If Alice wants to accept the requests, shecan take the requested information (PII or paymenthistory), encrypt it using the requesting party’spublic key, write it to IPFS, and attach the IPFSname to loan.

Payment Data Format

When DataCo shares payment information for auser, they are expected to share data in the fol-lowing format:

7

1 struct PaymentEntry {2 date: Integer3 amount : Integer4 }5

6 struct RepaymentLog {7 nonce : Integer8 Summary : String9 schedule : Array < PaymentEntry >

10 payments : Array < PaymentEntry >11 }

Consider the following data:

1 {2 nonce : 489376254 ,3 summary : "Cell phone bill payments

for 2017",4 schedule : [5 { date: 1483228800 , amount : 100 },6 { date: 1485936000 , amount : 100 }7 ],8 payments : [9 { date: 1483228800 , amount : 100 },

10 { date: 1484438400 , amount : 20 },11 { date: 1485936000 , amount : 100 }12 ]13 }

This represents an agreement that requires a pay-ment of 100 on January 1st 2017 and February 1st2017. The payments section indicates three pay-ments on January 1st, January 15th, and February1st. All past payment information is written toIPFS and a reference is stored on the user’s Bloo-mID. The summary should be displayed to Alicewhen asking her to confirm the payment informa-tion.

When a loan is paid off or the recipient defaults, theloan provider is expected to consolidate the loan in-formation down to this format and attach it to therecipient’s BloomID.

Tamper-Proofing Reported Data

When DataCo reports information to the loan con-tract, they share the IPFS uri of the encrypted in-

formation and the address of the intended recipient:

1 report ( address _recipient , bytes_ipfsUri ) {

2 // ....3 }

The recipient can interface with the contract toshare the data with other collaborators:

1 shareReport (2 bytes _originalIpfsUri ,3 address _recipient ,4 bytes _ipfsUri5 ) {6 // ....7 }

When an organization accesses a shared resourcefrom contract, it can lookup the original IPFS re-source to check that the original signature matches.The nonce helps deter against attackers generatingdifferent plausible JSON until one matches the sig-nature. A shareReport will be rejected by the con-tract if the sender and originalIpfsUri don’t matchan existing report. The loan recipient is never al-lowed to issue an original report to their own loan.

Fair Credit Reporting by Default

Bloom’s privacy model puts loan recipients at thecenter of all transactions involving their private in-formation and credit history. Users can review theinformation before sharing it with the company per-forming a risk assessment. In the event that infor-mation is incorrect, the user can work with the datavendor to amend their records using the same meth-ods available today. This workflow promotes proac-tive correction of information before it impacts auser’s BloomScore. Sharing data with the risk as-sessment company authorizes them to update theuser’s reliability score. Unlike traditional credit sys-tems, users can catch mistakes before they impacttheir creditworthiness.

8

When a loan has matured, the loan provider is ex-pected to consolidate the loan payment scheduledown to the previously mentioned format, publishit to IPFS encrypted for the recipient, and attachit to the recipient’s BloomID.

Risk Evaluation and Reliability Scores

Once RiskCo has received all of the payment infor-mation required in order to assess Alice’s creditwor-thiness, RiskCo is required to:

1. Update Alice’s reliability score to reflect anynew payment information attached to herBloomID

2. Approve or reject the loan

RiskCo does not have to make their decision forthe loan solely based on the BloomScore, but theyare required to update the BloomScore using theformula agreed upon by the network.

Alice’s updated reliability also triggers updates tothe peer scores of the users she has staked.

6. BloomScore (Credit Scoring)

The goal of the Bloom network is to securely exposeanonymized information about financial networksand historical payments so both lenders and bor-rowers benefit. The Bloom network will compute aBloomScore for each user that evaluates their cred-itworthiness.

We can think of a Bloom user as being in one ofthree phases of account maturity as she enters theBloom network:

1. The user has just recently signed up and hasonly staked other users, none of whom are fi-nancially active yet on the network

2. The user’s peers are financially active3. The user has taken out loans or otherwise has

financial information available in the Bloomnetwork

Bloom users in each phase are able to improve theircreditworthiness. Each phase corresponds to moreknowledge about the user and raises the ceiling forhow high the user’s credit score can be before theuser ascends to the next phase.

9

The initial Bloom release computes a simple scorebased on past debt obligations and payment his-tory. As the network grows with time we expectto increase the sophistication of the score basedon what usage best predicts outcomes. Proposedimprovements will be vetted and voted on by theparticipants in the ecosystem in response to howreal-world use of the protocol unfolds.

The initial Bloom score will be between 0 and 100,inclusive. Phases one and two will be capped at20 and 50, respectively. Each phase is scored dif-ferently in order to optimally fit the informationavailable.

Users will start with a score of 10. New users willthen stand out from users who have defaulted orscammed and therefore have a score closer to zero.

Scoring Phase 1

Users who have not staked any users with finan-cial activity will be scored on the number of stakesthey have established and how long it took themto establish those stakes. Individuals from more fi-nancially responsible social networks should be ableto find more people they can stake, so they shouldhave a higher number of stakes, and they should beestablished closer to the user’s signup date.

New users should be able to quickly get to a scoreof 20 if they add eight or more stakes in a week orless. Fewer numbers or longer time to acquire stakesshould reduce the number the user can reach. Thetarget number of users required to get a perfect 20will be lower as the Bloom network is starting off.

Calculating Reliability Scores

A reliability score is a prediction of whether a useris likely to pay back a future loan on time giventheir past financial activity. Bloom will start witha reliability score that takes into account:

• Total amount paid vs. total amount owed

• Longest repayment history on file

• Average payment total per month

• Number of past loans

• Total amount paid across all reported infor-mation

In order to make a prediction from these indicators,we will calculate a multivariate logistic regression.The indicators above would be represented as a vec-tor x = (x1, x2, x3, x4, x5). Each indicator will alsohave a corresponding weight such that our logit is:

g(x) = β0 + β1x1 + β2x2 + ...+ β5x5

and our regression is expressed by

R(u) = eg(u)

1 + eg(u)

where β0 is an offset, βi is the weight for the xi

indicator, and u is a user on the network. For thesake of having a simple implementation, each in-dicator will be bucketed into a discrete category.For example, ”total amount paid across all reportedinformation” will initially be included by takinground(log(total)) to reduce the variable down toa bucketing by order of magnitude.

The transformation of continuous indicators intodiscrete values requires adding dummy variables foreach indicator. So, for example, if xn has m discreteoutcomes then the actual calculation of βnxn wouldbe

m−1∑i=0

βn,idn,i

where each dn,i is a dummy variable for xn and βn,i

is the corresponding weight.

The final calculated score will be scaled up so thatit is between 0 and 100. The weights, indicators,and indicator categories will be subject to a voteon the Bloom network.

10

Scoring Phase 2

The peer score for a user will be the average Bloom-Score of each peer the user has staked, capped ata maximum of 50. In the equation below, s is thenumber of peers the user has staked.

min(50,∑s

i=0R(ui)

s )

Scoring Phase 3

In the final phase of account maturity, the user hasher own financial activity as well as financially ac-tive peers. Their BloomScore is now uncapped andthe weight of each payment they make weightedequally to the BloomScore of one of their stakes.In the equation P and T still correspond to “paid”and “total owed” but instead of P (u) we write, forexample, P2 to mean the second payment the userhas ever paid.

∑si=0

R(ui)s+n +

∑nj=0

Pj/Tj

s+n

7. The Loan Risk AssessmentLifecycle

Putting these concepts together, below is a simpli-fied example of the full risk assessment lifecycle in-cluding one identity attester and three credit stak-ers.

This example illustrates how one would porta traditional fiat-based lending lifecycle to theblockchain, but in less developed markets, the re-quired attesters and credit stakers could be cus-tomized to fit the demographic that is being evalu-ated.

1. Loan originator creates contract detailing theamount of the loan, the repayment schedule,and the requirements of the risk assessment

2. User reviews and agrees to the contract3. External attesters and stakers agree to verify

identity and creditworthiness based on loanoriginator’s set of desired attesters and stak-ers

4. Organizations update contracts with person-ally identifiable information (PII) require-ments that they need to fulfill attestationssuch as “name”, “address”, “date of birth”

5. User attaches encrypted details for each at-tester and staker to their BloomID contractusing the public keys of the respective parties

6. Identity verifier attests to the user’s provideddata

7. RiskCo pays alternative data provider (Telco)for anonymized transaction history

8. RiskCo uses additional payment history infor-mation to decide whether to stake user

9. Requirements all met. Borrower can with-draw credited funds

8. Bloom Token (BLT)

The Bloom Token (BLT) is both the currency andscoring enhancement mechanism of the Bloom net-work. The Bloom token allows organizations to par-ticipate in evaluating user identities and creditwor-thiness. It also serves as the proposal and voting

11

token to guide the evolution of the Bloom creditscoring system.

Network Currency

In the current credit infrastructure, it is commonfor lenders to pay identity verification and creditrisk assessment companies for their services. Sim-ilarly, identity attesters and risk assessors on theBloom network will be able to set prices and re-ceive payment for their services in BLT.

Distributed Scoring Enhancements

One of the primary functions of BLT is to serveas a proposal mechanism for instituting changes tothe BloomScore phases and algorithms. This pro-posal mechanism allows Bloom to maintain a creditscoring system that evolves according to the needsof its users. As identity and risk attesters providetheir services to lenders, they will accrue an amountof BLT roughly proportional to their benefit cre-ated within the ecosystem. This in turn will allowlenders, identity attesters, and risk assessors to pro-pose and vote on the changes they would like to seehappen at a credit scoring level. Our decentralizedscoring enhancement system will be handled usingAragon or some subset thereof.

Accrediting Attesters

Lenders and organizations within the Bloom net-work can have their customers’ payment historiesreflected in an individual’s BloomScore. This issimilar to how credit bureaus work with privatecompanies both big and small to form reporting re-lationships where nonpayment for loans is reflectedon an individual’s credit report. To handle thisin a decentralized environment, payment historyproviders and lenders will need to submit proposalsto the ecosystem regarding why they are trustwor-thy, what their business does, and why their datashould be included in BloomScores.

This will take an approach similar to the AdChainRegistry wherein BLT is paid by organizations ap-plying to be included in the BloomScore and users

who vote on that organization’s inclusion receive areward funded by the application fee.

Voting users will be required to review applicationssubmitted by these organizations for legitimacy todetermine if they should be included in the Bloom-Score by completing activities such as reviewing theapplicant’s website, researching the directors andother controlling parties, and reviewing the appli-cant’s proposed means of contributing to the net-work.

Bloom Invitation System

While the network is in its infancy without a widearray of identity attesters, lenders, and risk at-testers, it is more susceptible to attack. In orderto account for this in the bootstrapping days of thenetwork, Bloom will have an invitation system inwhich BLT users will be required to put a fractionalamount of BLT up as collateral for users who theyinvite. This is not collateral to ensure that invitedusers do not default on loans in one-off instancesbut rather collateral to serve as protection againsta mass-scale network of malicious accounts.

Requiring some BLT to be put up as collateral foran invite adds a negligible amount of friction forsigning up and inviting friends. The small amountof BLT required to create an account builds up ameaningful amount of capital if an attacker wants tocreate a network of malicious accounts. If membersof this network end up committing exit scams, thenthis upfront cost is mostly sacrificed, making exitscams via large fake networks more expensive evenduring the bootstrapping phase of the network.

Any BLT collateralized in the invitation system willbe returned to the sending users after one year.

This invitation system will also serve as a way ofconcentrating launches of the network within dif-ferent communities around the world. The Bloomsystem relies heavily on network effects and thus theinvitation system will help Bloom overcome the ac-tivation energy that would otherwise be prohibitivein a fully organic and potentially sparse networkspread.

12

9. Roadmap

The Bloom protocol will be developed in 6 majorphases:

Phase 1: Bloom Invitation System and Vot-ingPhase 1 will allow for users to use BLT to invitetheir friends and colleagues to seed the initial net-work securely. Users with BLT will be able to voteon early development-related proposals for the fu-ture of the network.

Phase 2: Bloom Identity Matching (Bloo-mID)Phase 2 will deploy an application allowing usersto verify their identity and get matched with theirBloomID. During this phase, users will be able toconfirm identity information and connect data forBloomIQ.

Phase 3: Credit Staking (Precursor toBloomScore)Peer to Peer staking modules will be built first, fol-lowed by organizational staking.

Phase 4: Creditworthiness Assessment(BloomScore)Phase 4 will allow users to check their score, aswell as open up a developer ecosystem to lendersto check a given user’s BloomScore, providing suffi-cient privileges are granted from the loan recipient.

Phase 5: Bloom Credit Protocol Launch +BloomCardOnce the risk assessment and scoring protocol iscomplete, Bloom will launch the BloomCard. TheBloomCard will serve as a brand new way for in-dividuals to display creditworthiness and improvetheir BloomScore.

Phase 6: Democratized Autonomous CreditInfrastructureOver time BLT flows through the network. Lenders,data attestation providers, and borrowers will allown Bloom Network Token and their amount ac-quired will correlate to their influence on the net-work.

Governance of the scoring protocol will be grad-ually turned over to BLT holders, granting themdetermination over:

• Which organizations will be able to updateand provide information to a BloomScore

• The weights and factors considered in theBloomScore

• Development and scoring protocol updates

10. BloomCard

To help expedite the adoption of the protocol,Bloom is launching the BloomCard. BloomCard isa blockchain credit card built on the Bloom proto-col. BloomCard is intended to serve as a model forall future credit providers and simultaneously allowthe Bloom protocol to be deployed and developedin a live environment. The BloomCard will start asan Ethereum debit card and additional functional-ity will be added over time.

BloomCard is the first project based on Bloomcredit infrastructure, launched alongside the Bloomprotocol. The intention is not for Bloom to becomea large-scale credit provider. Instead, BloomCardis an example that sets a precedent for other lendersand to help users build up a BloomScore in lieu oftraditional transaction data. Transaction data cre-ated by using the BloomCard that can in turn befed into the BloomScore.

Users whose application is supported by theirBloomScore will be able to obtain a BloomCard,whether they have a FICO score and an establishedtraditional credit history, or live in a market thatcurrently lacks credit services.

13

BloomCard is differentiated from any other tradi-tional unsecured consumer credit provider on themarket by allowing global access to credit, includ-ing in underdeveloped markets. Before launching,BloomCard will be compliant with relevant regu-latory standards and aims to launch first in theUnited States. BloomCard will use ETH as itsmeans of payment and will provide users with thebest-available spot exchange rate from ETH to USDwithout any unnecessary mark-up.

The BloomCard will be the first credit card builtentirely on the Bloom protocol. The BloomCardaims to offer similar incentives that users have cometo expect from the high-end credit card marketsuch as no foreign exchange transaction fees. Un-like a traditional debit or credit card, purchaseson BloomCard contribute to your BloomScore.BloomScore measures your purchasing power, fre-quency of purchases, and payment consistency tobuild your credit.

11. Team

Founding Team

Jesse LeimgruberJesse studied computer science at Stanford Univer-sity. Jesse is an advisor to The Alchemist Acceler-ator, a Thiel Fellow, and a mentor at the EuropeanInnovation Academy. He’s served as a guest lecturerat Stanford University, The University of SouthernCalifornia, DePaul, among others. Prior to Bloom,Jesse founded enterprise analytics software, Neo-Reach. NeoReach provides analytics for Fortune500 brands including Microsoft, Citrix, Walmart,among others.

Ryan FaberRyan Faber developed a behavioral recognitionmethodology designed to leverage online psycho-graphic data for user acquisition. Using his re-search, Ryan launched Flatiron Collective. Flatiron

now manages over $100M annually in digital mar-keting spend. His developments in user acquisitionhave allowed him to become a 3x Webby Awardwinner and his methodology has been attributed tothe exponential growth of numerous billion dollarbrands.

Alain MeierAlain Meier studied computer science at StanfordUniversity and served as a research scientist forStanford Bitcoin Group. Founded by 21 CEO, Bal-aji S. Srinivasan, The Stanford Bitcoin Group isStanford University’s blockchain research organiza-tion. Alain developed a number of open sourcecryptography projects including CryptoNote.me,an open-source service allowing users to send en-crypted, single-view messages in seconds. Followinghis work at Stanford, Alain is serving as the CEO ofcompliance and identity verification company, Cog-nito (formerly BlockScore).

John BackusJohn is a founding research scientist at StanfordBitcoin Group and studied computer science atStanford University. He is a Thiel Fellow and co-founder and CTO of identity verification company,Cognito. John is an expert at identity infrastruc-ture, previously engineering data preprocessing al-gorithms for large-scale entity extraction for deter-ministic and probabilistic record linkage. This iscurrently implemented into Cognito’s core identityresolution and record linkage infrastructure, nowprocessing identity and compliance for tens of mil-lions of cryptocurrency users globally.

Daniel MarenDaniel Maren studied computer science at StanfordUniversity. He founded solar power electronics com-pany, Dragonfly Systems, which was acquired bySunPower Corporation in 2014. Daniel remains anadvisor to SunPower, guiding solar + storage prod-uct efforts and international business development.He has expertise in international infrastructure de-velopment, finance, and energy, where he is a recip-

14

ient of a Forbes 30 under 30 award. Previously, hearchitected a biofuels program for Indonesian farm-ers, who struggle with seed and equipment financ-ing.

Shannon WuShannon is CEO of Mr.Progress, a leading market-ing firm for fast-growing Silicon Valley companiesworking on transformative technologies. She hasworked with Visa, Coach, Jawbone, Branch andothers. She is co-founder of Sourcerers.io, support-ing dominant Ethereum projects and blockchaininitiatives. Passionate about disruptive models,she has been investing, advising and partneringwith early stage companies for years. She fre-quently lecturers on marketing and company build-ing at Stanford University, her alma mater, andaccelerators. She previously headed marketing atFOUNDER.org, an early stage fund investing inyoung innovators.

Advisors

Joey KrugJoey is the Co-Founder of Augur, the world’s firstdecentralized prediction market and oracle system.Joey has been a pioneer in the Ethereum commu-nity for many years, having launched the very firsttoken project built on Ethereum. More recently,he’s advanced the cryptocurrency community in hisrole as Chief Investment Officer of Pantera Capital,a $100m ICO fund investing in token sales.

Luis CuendeLuis is the project lead and co-founder of Aragon,an Ethereum-based project that enables dApps torun entire organizations using the blockchain. Luiswas recognized as the best underage European pro-grammer in 2011, is a Forbes 30 Under 30 andan MIT TR35, and was an Advisor to the VPof the European Commission. He co-founded theblockchain startup Stampery. Prior to foundingstartups, he created the world’s first Linux distri-bution with facelogin.

Geoffrey AroneGeoffrey served as the Chief Scientist for Expe-rian Consumer, and was the SVP of Product In-novation for Experian Global. Experian is one ofthe “Big Three” credit-reporting agencies, along-side TransUnion and Equifax. It collects and ag-gregates information on over one billion people, em-ploys 17,000 people, and generated $4.6 billion inrevenue in 2016. In addition to serving as ChiefScientist and SVP of Product while at Experian,Geoffrey also served as Experian’s VP of MarketingStrategy and Innovation. He developed consumerforensics and insights, distribution strategies, andmarket expansion strategies.

Meg Nakumura, CEO of Shift PaymentsShift developed the Shift Card, a VISA debit cardthat currently allows Coinbase users in select statesand territories in the U.S. to spend bitcoin any-where VISA is accepted.

Joseph Urgo, Co-Founder at District0x, CIOat Sourcerers CapitalJoe is the co-founder of District0x, an EthereumdApp decentralizing the world’s marketplaces.Prior to this, Joseph founded Sourcerers.io, aconsultancy supporting leading Ethereum-basedprojects. Joe previously spent three years as anOperations Manager for Coinbase. Prior to Coin-base, he was a derivatives trader for Three ArrowsCapital, an international hedge fund based in Sin-gapore.

David RaphaelDavid Raphael is the CEO of Infinity Media, a dig-ital agency specializing in conversion rate optimiza-tion. He studied at the University of Chicago andhas spent the last six years sharpening the product-marketing funnels for companies like Artsy, Fan-Duel and Wag!

15

Devon ZuegelDevon Zuegel studied Computer Science at Stan-ford University, where she also served as the Editorin Chief of the Stanford Review. She’s worked asa software engineer at Affirm, a financial servicescompany that provides an affordable, flexible, andtransparent credit option. Devon built out Affirm’s

identity infrastructure, ensuring that users are whothey say they are and linking that information intothe credit underwriting system, which funds mil-lions of dollars of loans every day. Now, she worksas an independent consultant advising startups ontechnology and strategy.

16

Special ThanksA very special thank you to the people who helped to review this whitepaper:

• Spiros Zarkalis• Stelios Manolopoulos• Devon Zuegel• Daniel Maren• Brian Sorel• PJ Leimgruber• Shannon Wu• Joe Urgo• ray-jones

References

[1] https://royce.house.gov/uploadedfiles/credit score competition act.pdf

[2] http://www.myfico.com/credit-education/how-lenders-use-fico-scores/

[3] https://www.consumerfinance.gov/about-us/newsroom/cfpb-report-finds-26-million-consumers-are-credit-invisible/

[4] http://documents.worldbank.org/curated/en/187761468179367706/pdf/WPS7255.pdf#page=3

[5] http://www.perc.net/wp-content/uploads/2013/09/web layout-you-score.pdf

[6] http://pubsonline.informs.org/doi/abs/10.1287/mnsc.1120.1560

[7] http://pubsonline.informs.org/doi/abs/10.1287/mnsc.2015.2181

17