 is a financial modeling platform powered by reliable on-chain data, currently built for Ethereum, but eventually expanding to other EVM-compatible chains. Credmark's Terminal aims to be 

 source of Risk Data in the DeFi world. Powered by Credmark's ongoing research and open-source 

, the Credmark Terminal will be continuously updated as risk reports are verified and validated.

Credmark has solved many of the challenges of on-chain data transformation, cleaning, and normalization. As most data scientists and quantitative practitioners know, the majority of their time is not spent building models, but rather normalizing and cleaning data. Even raw blockchain data is neither optimized nor immediately ready to build models and needs some transformation. Credmark has completed this process which allows a data scientist or modeler to building immediately. Here are 

 with some of the many data problems associated with building models on blockchain data that Credmark has solved.

The Credmark protocol uses a combination of cloud and blockchain services to create, test, and execute data models on a normalized, curated dataset from distributed providers. It then allows a variety of mechanisms to consume the outputs of the models, configured by a set of decentralized governance mechanisms.

In order to maintain long-term stability, governance will determine additional datasets to be added to the platform. Credmark’s data standardization layer abstracts the problem of unique data sources providing disparate APIs which do not deliver normalized data. This normalized data can then be added directly to the Credmark API.

In order to produce consistent, high quality data and risk models with sufficient predictive capacity, it’s a necessity to have high quality normalized data that is open to everyone. Without this high quality normalized data, model contributors spend an excessive amount of time focused on optimizing models with subpar results.

Credmarks Financial and Risk Modeling Advantage

In a few words: it’s Python and open source.

, or other on-chain analytics platforms, except Credmark differentiates itself by natively supporting Python for creating open source models that any modeler can build on. Compared to many of the on-chain analytics platforms which are proprietary - the data, analytics, and models cannot be verified and validated.

All blockchains are essentially large, poorly optimized open source SQL databases, hence the reason that all other on-chain analytics platforms like Dune are built on SQL, because it’s quick and easy.

However, from a risk and financial modeling perspective, SQL will only get the blockchain and crypto industry so far. SQL is designed specifically for querying large databases and creating analytics, but it is not designed to build risk and financial models.

As the crypto industry matures and growth and adoption has dramatically declined, risk and financial modeling likely will become the differentiator for blockchain protocols that survive and thrive the next several years. This transition to move beyond SQL will be a necessity to perform proper risk modeling and management that can only be built in languages like Python or R that are designed to build sophisticated models.

Below is an example diagram of the maturity of data science modeling within an industry/organization and the resulting analysis and data driven decisions that can be derived. Starting from operational decisions, up to strategic and transformational data driven decisions based on predictive analytics and machine learning automation.

Our classification would put the majority of the dashboards and analytics that exist for crypto today are generally operational and KPI reporting analytics. To transition into strategic and transformational data driven decisions, protocols will have to develop more sophisticated risk and financial models in Python or similar languages.

This is a massive advantage for Credmark since the protocol is built on Python, especially given all the thousands of native financial, data science, and machine learning algorithms and libraries that can be leveraged in Python. These algorithms simply don’t exist for SQL. If coded appropriately, this could also give Credmark an execution speed advantage.

Irulast plans to take advantage of this infrastructure and will use Credmark for much of our blockchain DeFi risk modeling on Ethereum. Irulast has recently completed a proof of concept building some DeFi risk models on their platform for some stable coin risk analysis. The models interacted directly with on-chain data and we’re excited about the protocol and its applications for risk and financial models.

 takes you the last mile, delivering your software to production.

We offer custom application and system development services from seasoned developers featuring expertise in blockchain and smart contracts, application development, systems architecture, and DevOps.

How Irulast is Leveraging Credmark | Guest Post

Data Modeling on Credmark

July 2022 won’t be remembered as a fun month by anyone in the crypto community.

In June we lost a stablecoin protocol that kickstarted a chain of events that led to the collapse of not a few large centralized crypto institutions. It’s ugly. The only winners – and this is going to be important – are DeFi lending protocols and DEXes. They have proven extraordinarily resilient.

Despite these successes, no protocol or project has avoided the impact of the centralized meltdown. Treasuries have been decimated, investment money has been pulled, and many people have lost their jobs.

Some projects have already shut down, some will shutdown soon, and some are restructuring to survive in these conditions.

Credmark falls into the third category. Our treasury is in relatively good shape, but we’ve shut down a planned fundraising effort due to unfavorable market conditions. We are now making changes to ensure our future success.

My co-founder, Neil Zumwalde, has been a fantastic technical CEO. He ran the company during a critical building period. Under his leadership we developed the best data platform in DeFi. This platform is complex. Unfortunately, selling it requires more education and time than we have.

Because of these constraints, and because of my background selling financial services products, Neil asked me to lead the company through a technology consolidation and early sales phase. We both believe that this is necessary before re-engaging with investors.


Unlike many DeFi projects, we have customers and revenue. We will continue to service these clients and to sell access to the platform we developed. Their feedback has been invaluable. It has allowed us to design more focused, simpler API products. These will be announced over the coming weeks. When we deploy our protocol, we will support anonymous API access. In the meantime we will allow customers to purchase API keys with credit cards.

 will look more like a traditional SaaS business than a community-driven data and modeling platform.

Because we have delayed the introduction of our protocol, our $CMK token cannot provide its anticipated utility. We have therefore decided to sunset it.

If you are a $CMK holder, this is what you can expect.

The token’s reported value will drop to a fraction of a penny, and the token will be delisted. $CMK and $xCMK (staked $CMK) tokens will not, however, disappear from your wallets.

We fully intend to launch our protocol when resources and market conditions allow. At that point we will either provide a mechanism to convert your current tokens or airdrop a new utility token. Our core team has discussed this in depth, and we are excited about future possibilities. In the meantime we will focus on continuing to build the world’s best DeFi data platform.

We appreciate the support of our early token holders, and we thank you for sharing our vision.

 and reach out to any of the core team members. Our mission hasn’t changed. July has been painful but we’ve experienced adverse market conditions before. We will survive and deliver the platform we promised.




A New Beginning

The End of an Era. The Start of Another.

Information in DeFi and the wider crypto industry is widely accessible but scattered and often incoherent. It’s certainly not an understatement to say we’re dealing with information overload. To further complicate the situation, KOLs with crypto-cult-like behaviors endorse a project or shill tokens, and then casually throw in a DYOR (do your own research). It's suffocating and difficult to ascertain what’s true. Needless to say, access to reliable data and insights requires intensive research, careful analysis, and the ability to synthesize many sources of information. Instead of placing faith in individuals, well-informed decisions should be provable, backed with reliable on-chain data, and rooted in quantitative sciences. There’s a market for DeFi insights backed by clean data and 

It all started with the Risk Terminal which was launched last December as an MVP. We released a dashboard that was built on a few basic models we’d developed. Visualizing these models’ output was powerful. In the meantime, our community built more and more models. So we took a step back and wondered if we could coherently organize them. This effort led to the release of what we now call the Credmark Terminal. Behind the scenes, we do all the number crunching so users can see the outputs of validated financial models like 

. (Learn more about why we started with these models 

.) In a sense, the Credmark Terminal is a showcase for our models. All of the code is open source so anyone in the community can grab it to build their custom dashboards. Over time, we hope the Terminal becomes a primary source of risk data and financial data in DeFi for people who don’t require access to the raw data available via our API.

The first models we’re deploying in the Terminal cover lending protocols like Aave and Compound. Their source code can be viewed 

. All financial models are validated and verifiable, so users can verify how the models are built. Rest assured, all data is sourced from Credmark’s reliable on-chain state and ledger data.

Although risk continues to be our focus, our community has asked for other types of financial models. We’ve begun supporting those and, by necessity, had to rename the product. That’s how The Risk Terminal became the 

We’ve implemented lots of feedback from our users, added a ton of exciting features, and redesigned the entire terminal. This new interface enables us to easily scale and include more exciting features as our model repository grows. Selected models will be displayed on the terminal.

 will showcase established TradFi metrics like 

 but tweaked for DeFi. Other protocol metrics like VaR, TVL, and Trading Volume will be added subsequently in the coming weeks.

 will feature metrics like VaR and LCR for Aave and Compound. The new PowerBI Dashboard will also show data on Aave and Compound such as the flow of assets, top 10 transaction count, top 10 wallets, etc

 cover VaR, TVL, and Trading Volume metrics for the Top-10 pairs across Uniswap, Sushiswap, and Curve

 will show all relevant stats on Credmark’s native token $CMK and display the models that are most popular in terms of queries via our API.

All of the data in the Terminal will be available on 30-day, 60-day, or 90-day timeframes and can be downloaded in CSV format.

Sharpe Ratio is an established metric to compare risks and returns of assets. As we adjusted it to be compatible with DeFi, users are now able to compare the Sharpe Ratio for up to 5 different tokens across a time series.

While the lender's section only covers Aave and Compound at the moment, the purpose of these charts is to showcase the liquidity and solvency of lending protocols over time. These metrics are useful indicators to assess when the protocols may be undercollateralized or how much value may be at risk in a black swan event.


The Lending Pool Usage is a great tool to see the top lenders or borrowers of DAI, USDC, and USDT stable coins on Aave and Compound. Users can quickly visualize:

Users can also perform more in-depth analysis by right-clicking on a dataset and pressing drill through. All information is neatly presented up to the exact transaction hash and block timestamp.

Providing liquidity is a lucrative way to earn money, but it is risky. Our dashboard can provide insight into potential risks with the use of VaR or inform users of the imbalance within various Curve pools.

Take for example, the USDC / ETH pool on Uniswap V3 has a VaR of -30% which indicates the possibility of losing up to 30% of your total LP position in a worst-case scenario.

Now if we look at the Curve 3Pool, the balance ratio less than 1 which reflects an  imbalance towards USDT in the pool:

The Credmark Terminal is an application that demonstrates how powerful Credmark’s data is. As more sophisticated models are built, the more utility the Terminal will have. A full list of the current models can be found in the 

. A lot of the current models are built with the same parameters as traditional finance. At Credmark, we’ve been at work to develop crypto and DeFi native models. This is just the beginning. We will release some useful interactive dashboards very soon. 

 and let us know what you think or what you would like to see.


 is a financial modeling platform for DeFi, powered by verifiable on-chain data. It provides the infrastructure necessary to build models. An API allows anyone to consume the output of these models. 

Our modeling tools are the most flexible and robust available today. The 

 streamlines prototyping and deployment. 

 members are incentivized to build and improve models. As a result, users benefit from verifiable data and best-in-class analyses.

Credmark is a decentralized project. Our community participates in governance, provides research, and develops models. 

 to advance the next-generation financial system.

A DeFi Source of Truth

Introducing the Credmark Terminal

UST, the largest algorithmic stablecoin, was under pressure for the last two days struggling to fulfill its core obligation, a peg to 1-to-1 USD. Reasons are both market and idiosyncratic events, which led to a spiraling crash. The market event was BTC trading breaking the side trend that potentially caused a lot of liquidations. Idiosyncratic event – withdrawals from highly concentrated UST holders.

By far the biggest on-chain DeFi operator in UST is Anchor, lending protocol with TVL equal to $8.5 bln. as of 10 May. UST withdrawals started on 7 May: $2 bln. on 8 May, $1 bln on 9 May and $7 bln. on 10 May. In total deposits dropped more than half from $17 bln. to $6 bln.

Onchain there are three biggest DEXs operating in UST with the following characteristics as of 10 May 2022:

· Curve TVL $0.8 bln. 24H Volume $2.1 bln.

· Uniswap TVL $19 mln. 24H Volume $218 mln.

· Sushiswap TVL $0.5 mln. 24H Volume $4.5 mln.

Curve has by far the largest activity in UST of which the majority belongs to UST +3Crv pool. Its balance in time shows that the balance of UST in the pool almost doubled, which confirms that there was a large exit from UST accompanying unwinding lending positions mentioned in the previous section between 9 and 10 May

Tracking major UST transactions (>$10,000,000, that represents around 30% of the whole volume) for the same time shows that the highest concentration of whales exit was through Abracadabra ecosystem

Over the same period UST withdrawals from Uniswap Pools was about $68 mln., of which $17 mln. was a volume from MEV arbitrageur. $40 mln (60% of the total volume) was sold over 7 hours from 4 to 11 am.

UST is a stablecoin and LUNA is a governance token that acts as a treasurer to maintain the obligation of UST to be pegged to $1.

Price of UST initially dropped to $0.975 (2.5% depeg from USD) on 9 May 2022 and then to $0.67 on the next day (33% depeg from USD) on 10 May. Total net supply grew from 351 mln to 801 mln tokens which confirms large structured positions unwindings (like staking on Anchor). LUNA was selling collateral and burning UST to drive the price up but the effort wasn’t sufficient, hence net supply doubled.

On-chain data analysis shows that most of the UST transfer volume on 9 May was minted further confirming large staking positions liquidation.

Luna’s treasury activities to maintain the peg further spiraled the market drop. BTC dipped due to holdings liquidations.

In line with expectations Terra Treasury activities affected the token price of the LUNA.

LUNA also experienced large volumes of DeFi positions liquidations accompanied by a selloff.

Although the majority of the transactions are occurring off-chain, DeFi data could be used as a good early alert system for protocols due to its transparency and verifiability. Large outflows and concentrations can be monitored and tested historically which can help in building a robust LQ risk management system and comparing different shops. LUNA from what is shown underestimated the amount and intensity of withdrawals resulting in a platform to be on the brink of existence.

All the data in the current report was sourced from the Credmark Model Framework as part of risk metrics to empower effective liabilities (in this case to maintain the peg to USD) and risk management.

What caused UST to de-peg

Last week, the biggest and most popular algorithmic stable coin UST de-pegged from USD 1 by 33% in one day followed by a death-spiral sell-off and now it’s trading below $0.1 after slightly more than a week. Its governing token, LUNA crashed by nearly 100% and subsequently was delisted from many exchanges, including FTX and Binance. It’s important to understand how this happened, backed with data and the application of quantitative methods

Algorithmic stablecoins are, in simple words, an automated central bank that, via built-in smart contracts, tries to control the supply and demand of the token so that its price is traded within the required range (peg). 

Terra’s UST was by far the largest algorithmic stablecoin ($29 bln. market cap on 5 May 2022) before its crash. It was running a so-called seigniorage model, without fixed collateral, where the peg relied on the arbitrage between UST and Terra’s native token LUNA. The price of the LUNA token reflects the development of the underlying ecosystem and applications, which includes the mining network. 

The design of Terra’s stablecoin UST was relatively simple, 1 unit of UST was backed by $1 dollar worth of LUNA, the governance token of Terra. Whenever someone exchanges 1USD for 1 UST, $1 dollar worth of LUNA was burnt. 1 UST can be then swapped anytime to $1 worth of LUNA. When UST is priced over the pegged of $1 dollar, one can mint 1 UST by burning LUNA and sell UST on the open market, and increase the supply of UST to bring it back to the peg. When UST is below the peg, one can mint $1 of LUNA by burning 1 UST and selling LUNA on the open market, and decrease the supply of UST.

There were many algorithmic stablecoins launched in the history of cryptocurrencies [1], but only sUSD, UST, and FRAX managed to maintain the peg of $1. DAI is conceptually different as it’s fully backed by third-party assets and is a decentralized stablecoin. 

The following table provides a breakdown of the size characteristics of UST, FRAX, and sUSD in $ bln as of 5 May 2022.

*Given each algo stablecoin is designed to maintain the peg to $1, the market cap of the stablecoin is considered to be a liability from an issuing platform perspective

** Self-reported, the information couldn’t have been verified by the third parties [2], [3]

The first UST de-peg happened on 7 May and resulted in the drop of LUNA’s token price and subsequently dropped l to ground zero by 13 May. The table below outlines the impact on the stablecoins market cap at the time before the crash and in the end in $ bln. As we can see, although circulation of non-UST algorithmic stablecoins decreased, they maintained the peg and their market cap remained relatively stable compared to the 92% ($17.23 bln.) drop in UST.

Hierarchy chart below outlines the biggest parts of the Terra ecosystem as of 5 May 2022. The arrows represent the direction of claims in case of withdrawal.

Here’s a breakdown of primary DeFi operators in UST and their holdings (Wrapped and Wormhole version in Eth main network).

Anchor is a lending protocol specializing in UST. It was known for its staking program which paid out 20% APY on deposited UST. The total deposits accounted for 75% of the total UST in circulation and paid approximately $0.23 bln. in monthly interest. There has been a large outflow of UST from Anchor (14bln UST before the crash and 1bln UST now). In the initial de-pegging of UST, holders were willing to sell UST at a 10% discount as they have cashed in the interests in the past months. The sale may have replenished the supply of UST in the pool and could have contributed to the further de-pegging of UST.

As UST leverages the capital which backs LUNA, it’s critical to assess whether the protocol is well-capitalized to ensure survivability under a stressed situation. In this regard, there are two risk management questions to answer:

What should have been the required capital-at-risk?

How was UST positioned against other algorithmic stablecoins before the attack?

2.1 How much was required to break the peg?

During the initial wave of withdrawals, on 10 May the UST dropped to 0.67 cents. This is a de-peg of 33% from the $1 set by Terra.

To model the amount required to break the peg by this much, we can use an Automated Market Maker (AMM) Decentralized Exchange (DEX), where the pricing is algorithmic.

Let’s assume for simplicity that it was a Uniswap v2 UST/DAI pool. The pricing, in this case, is governed by:

                                                                           x∗y=k

Where x and y are the quantities of UST and DAI, respectively, in the pool, and k is the constant number. We assume that both quantities are equal to $19 bln. worth. That is equivalent to the market cap of UST before de-pegging.

With the pricing algorithm, we can estimate that it would take approximately $3.4 bln. of DAI swapped against UST from the pool to move UST/DAI price from 1 to 0.67.

This is larger than the $3.29 bln. that LFG had, which might explain why UST only rebounded to 0.94 for the rest of the day before it crashed.

Liquidity coverage ratio (LCR) for algo stablecoins is recommended to be parameterized as:

Stressed liquidity buffer assumes a 30% reduction in its market value, and

Net outflows assuming the need to repay 25% of coins that could be exchanged to fiat within a 30-day window.

With the described approach we could compare UST and FRAX on 5 May 2022, 2 days before the beginning of the UST de-peg:

As shown, UST was well under-reserved. Its LCR was at 0.49 compared to 2.38 at FRAX.

Recommended level of LCR is at least 1.1 which means UST should have had a minimum liquidity buffer of more than double or two times fewer on-demand liabilities.

Since market volatility in cryptocurrencies is significant, market risk capitalization is one of the primary considerations.

As mentioned in section 1, Terra had a separate organization tasked with forming a reserve to support Terra Luna in fulfilling its obligations, LUNA Foundation Guard (LFG). Based on what the LFG disclosed on Twitter, as of Saturday, May 7, 2022, the LUNA Foundation Guard held a reserve consisting of the following assets: 80,394 $BTC, 39,914 $BNB, 26,281,671 USDT, 23,555,590 USDC, 1,937,554 AVAX, 697,344 UST, and 1,691,261 LUNA. for its treasury as of May 7, 2022, before it started to sell in an attempt to pull UST back to pegging, we calculate its VaR (180 days horizon, 10-day return, 99% confidence).

As of 5 May 2022 VaR was $0.52 bln. and represented 17% of LFG reserve. VaR remained stable around the $0.5 bln. mark and didn’t increase, which could mean that the volatility observed over the period between the beginning and end of the crash was in line with Business-as-usual volatility observed in BTC, the largest part of the LFG reserves.

FRAX runs a much more conservative model with 85% of FRAX circulation stored in USDC as collateral and 36.6m$ in various altcoins with the same 10day VaR equaling 0.

As of 5 May 2022 required market risk capital in $ bln. per protocol are:

Market risk capital should have been set aside to cover the risk of market volatility.

Terra UST had flaws in their treasury (assets and liabilities) management:

Community relationships - high concentration of investors and lenders in decentralized protocol

Product - all liabilities were effectively on-demand without any term structure making it an easy target for a bank run

Tokenomics - a system of incentivization was relying on the assumption of growth and neglected other instruments such as UST staking APY that could have disincentivized sell-off

3.1 The concentration of investors and lenders

Was it an attack from one concentrated holder? Yes, it was, we’ve identified significant withdrawals from Anchor made by individual wallets starting from 7 May to 11 May, with the largest made on 8 May that was $347 mln. worth of UST. Pie charts below clearly show this type of behaviour and overall UST inflow was toppled by UST outflow.

UST Liquidity added/removed (deposit and redeems) to Anchor protocol From 5 May - 12 May:

Interestingly, the Top 10 addresses were responsible for around 30% of overall UST liquidity removed during the period, which is very high considering best practice concentration shouldn't exceed 5% per counterparty. The daily break-down of redeeming for the top 10 addresses is below:

Here are stats of the total deposited and redeemed UST in Anchor during 5-12 May.

In summary there were $8.5 bln. net outflows between 5 and 12 May from Anchor.

Analysis of transactions from the Terra chain confirms that during the first wave of redemptions there was an outflow from a concentrated number of UST depositors equal to 2.5 bln representing 18% of the entire balance and 30% of the whole withdrawal over the week from 5 May. The holders concentration level of more than 5% is considered to be very large. It should have been better managed by Anchor treasury management team through diversification or restructuring.

Also since Anchor had only one staking product which was on-demand, in other words, depositors had the right to withdraw at any time. The term stacking should have been introduced to spread out cash outflows.

Anchor as the biggest lending protocol played a significant role in Terra UST incentivization mechanism. The rate that was offered to lenders was kept constant at 20%.

Firstly, the rate offered was unsustainable given the size and concentration ($17bln), staking could have been diversified at smaller APYs across different platforms organically. Terra had to find investment opportunities with higher APY to maintain a positive margin.

Secondly, in an unprecedented bank run, the rate of borrowing remained static. Whereas it could have been used as a tool to incentivize inflows and maintain the peg. In particular, introducing a variable rate that would change dynamically to support the peg algorithm. Or introducing term structures to incentivize stickier behaviour.

From the lagged nature of actions (the chain was not halted until11 May after 4 days of sell-off). It appears the Contingency Funding Plan lacked robustness and authority to act. This is a typical reason for failures in TradFi institutions before the global financial crisis.

One particular one that is worth mentioning is the ability to timely shutdown technology that might have a flaw that can lead to liquidity crises.

Data and financial modeling help to extract rational insights on DeFi protocols. Our analysis shows that both risk and treasury management activities required improvements.

Counterparty and maturity concentration thresholds weren’t properly managed. Incentivization tooling was sub-optimal and lacked a variety of tools. The asset and liability management function lacked robust liquidity crisis management expertise and authority.

UST was a high liquidity risk model proven by LCR metrics. LFG should have had separate capital to cover for potential market risk events given its reserves were placed in risky assets. Risk metrics should have been better aligned with peers.

Coingecko, How to DeFi: Advanced, https://www.coingecko.com

LFG, accessed on 16 May 2022, https://datastudio.google.com/u/0/reporting/b31cc9e5-c54c-4418-a6ce-b332c57e82e9/page/4YBqC?s=or-T7NeGLew

FRAX dashboard, accessed on 16 May 2022, https://app.frax.finance/amos

LFG Twitter account, accessed on 16 May 2022, https://twitter.com/LFG_org/status/1526126703046582272

Terra’s UST post-mortem, risk and treasury management

Simply access verifiable on-chain state and ledger data using 

Blockchain data is public and immutable. Smart contracts are also public, so everyone can see what they do. This transparency and immutability lead people to conclude that blockchain data can be trusted. It can and it can’t be. Almost nobody consumes the raw data, which is reliable. Instead, we consume interpreted data, and that interpretation is not reliable at all.

Web3 participants are unlikely to challenge the validity of the data they consume even though they know nothing about how it was produced. Data collection and interpretation methods are either obscure or unknown. Luckily, this will gradually change as the industry develops and the race to acquire higher quality data accelerates. More competition will inevitably lead to more stringent demands for reliable data.


On-chain data is raw when it is first extracted from the blockchain. It has to be interpreted and indexed before it can be used. In DeFi, this data is produced by smart contracts, so the interpretation of the data depends on the contract (program) that produced it. There are millions of smart contracts deployed on the Ethereum chain alone. These multiple data sources complicate things further. A closer look will present 

 like how on-chain data timestamps and real-time data like oracles and API timestamps present different ways of expressing time.

Deriving insights from data require careful analysis and data normalization as well as consistent methodologies. All of these must be transparent and verifiable so users can now rely on them.

Web3 and DeFi are touted as trustless because of their transparency, but to what extent is this true? Most providers of data or DeFi protocols do not fully disclose contextual details.

The truth is, it’s all a black box and the methodology isn’t transparent. Say you want to fetch the number of swaps that were conducted using the protocol’s API.

How do you assess the accuracy of the output?

These are all good questions and we can’t answer them without full transparency. In the absence of this transparency, we assume the data is correct.

To showcase how discrepancy exists across multiple reliable data sources, we conducted 

where we queried liquidity data across three sources for the lending protocol, Aave V2.

Smart contract (on-chain) data extraction

On-chain analysis is conducted by measuring the amount of aTokens issued by Aave to depositors. The difference is between assets' balance fetched from Aave Native API and total supply of corresponding aTokens. 

 From the comparison, a major difference is present in the total number of assets (totaling up to several million in value).

On-chain was point-of-time before block creation

Subgraph data can be collected at any time of the data with no clear transparency on how the node operates and the variables it takes into account.

Different sync times with variance across outputs can be troubling. Not to mention, the outputs will be difficult to verify and reproduce. Check out the 

Data availability is crucial when we consider the amount of value transacted every second in crypto. Imagine if a rebalancing application cannot rebalance swiftly because:

the source is slow or unresponsive, or worse,

Users of more reliable data can react better and faster to market conditions. Reliable data is worth money. Unfortunately, these are common issues that persist in DeFi. Luckily, they are being addressed by a new generation of independent data providers.

These are a few of the problems we’ve encountered. It’s not to suggest that these are the only problems.

At Credmark we work publicly and transparently. All of the data we produce can be traced back to the original, trustworthy on-chain data. This data drives our finance and risk models that in turn produce data. All of this is available via our API.

Our engineering and modeling teams are always available. They respond to queries and process community feedback as quickly as possible.

To accelerate the development of decentralized finance, access to data with verifiable integrity is essential. Together, we can do this.

 to advance the next-generation financial system

A Primer

Data Problems in DeFi

I met Wen-Chiao in May of 2019. Credmark was 5 months old. We’d just gone through an accelerator program in Hong Kong.

For some reason – I really can’t remember why – I was in Tokyo. The accelerator’s Program Manager sent a note telling me that a friend of his wanted to meet. I had no idea what to expect, but I agreed.

Wen-Chiao suggested a coffee shop deep in the bowels of one of Tokyo’s train stations, the kind of place foreigners leave their luggage in lockers they’ll never find again. After some trial and error, I found the coffee shop, and Wen-Chiao.


When we started talking I realized he was unusual.

His speech gets faster as he finishes his thoughts, as if needing to catch up with his brain. He frequently punctuates his sentences with a breathless ‘yea’, which sounds like a confirmation, or a victory cry. I’m not sure which, but I love it.

Wen-Chiao sounds American. I assumed he was. Then he ordered a drink, and I thought he might be Japanese. When I asked him which it was, he laughed, and said: “I’m Taiwanese, yea.”


After that unusual ice-breaker, I asked Wen-Chiao why he wanted to meet. He told me he loved crypto and wanted to work with us. When I asked him what he wanted to do, he said: “I’ll do anything. I know a lot of people in crypto in Japan and Taiwan.” That sounded great. We needed to find potential customers in Asia. Only problem was, we couldn’t pay him. The best I could offer him was a bonus on success. Wen-Chiao thought that was fine and soon disappeared into the station.

Wen-Chiao is tenacious, and creative. We talked every week, and every week he had something new to tell me. He came up with all sorts of leads and introduced me to many people. But commercially, we made no progress. We were too far ahead of the market.

Wen-Chiao busted his tail, but never earned a dime of commission.


A year ago I told my co-founder, Neil, that I couldn’t continue as CEO. We weren’t making any progress, and I wasn’t capable of restructuring the company to address a more immediate problem. Luckily, Neil was. He took over and turned Credmark into what it is today.

Leadership changed. Our mission changed. Our product was ditched in favor of something a lot more ambitious. Despite the turmoil, Wen-Chiao stuck with us. He saw what Neil was doing, and he thought it made sense.


One day Wen-Chiao called me and said: “I’m going to disappear for a few months. I never finished my military service, and my time’s running out.” Disappear he did, for 6 months. The only time I heard from him was during short home leaves. He stayed in the loop.

During this time, we raised money and started building a robust organization. Just as we started looking for someone to manage our growing community, Wen-Chiao was released from the Air Force – forever 🤞🏻– and I suggested to our then Head of Marketing that he should talk to him. He did, and a week later Wen-Chiao was a paid consultant.


Wen-Chiao took to his new job like duck to water. He learned all the DAO tooling, managed our English community, our Chinese community, and our Japanese community. As far as I could tell, he never slept. In a company of hard-working oddball polyglots, Wen-Chiao fit right in.

Six months later Neil decided that our Head of Marketing should be responsible for growth. I had to find his replacement. I wondered: could Wen-Chiao do it?

I hired a first-rate marketing consultant to help me solve the puzzle. She worked with Wen-Chiao for a few weeks before letting me know that he was definitely Head of Marketing material. He asked the right questions, thought about the right issues, and worked hard to fill in his lacunae. So we gave him the job.


People love to talk about the opportunities available in DAOs. “All you have to do is show up and contribute.” In some cases, it’s true. Wen-Chiao showed up. He believed in Credmark through two different iterations. He stuck with us, and now he’s a critical part of the core team.

I couldn’t be happier with this outcome. I’m a little bit proud of the role I played in this journey even though 99% of the credit goes to Wen-Chiao. I’m looking forward to supporting him in the coming years.

The message to anyone else interested in joining our community is clear. We welcome contributors. If your work is valuable, you’ll be rewarded. We’re paying attention.


From “I’ll do anything” to Head of Marketing

Meet Wen-Chiao

How to install and run the Model Framework

 as an open beta.
We’ve received a lot of positive feedback and comments from a wide range of users like:


 if you haven’t. It’s a game-changer and we’re setting on a mission to help users build robust models, deploy applications faster and access the most reliable data.

how the Model Framework can greatly reduce lines of codes 

 to give things a spin, provide feedback and help us iron out any issues.

You get rewarded to build models, propose improvements, find issues, and help spread the gospel.

Credmark Blog

For the latest technical analysis, partnerships, and community updates