How to install and run the Model Framework

The process of fetching reliable DeFi and crypto data isn’t clear-cut. Say you want to know the historical price of a token, you’ll probably look to centralized services like CoinGecko or decentralized alternatives like SubGraphs or others like Dune Analytics. That’s all fine until the validity of the data is brought into question:

How would you know when or how the data was collected?


How would you know the exact block of the transaction?

The truth is, it’s all a black box and the methodology isn’t transparent. Reliable models require reliable data, precise context, and verifiable sources. 

This is where the Credmark Model Framework comes in.

, users can make use of existing and verified on-chain data, but it abstracts away the complexity of web3 coding. 

Input the block number or the time-stamp to output the transaction data. If there’s no data for the exact time-stamp, it’ll provide the closest data to the input time-stamp and corresponding block number. 


Don’t have the bandwidth to look for all the individual smart contract addresses? Input the smart contract address of the protocol, and it will output all the contract’s ABIs. 

It simplifies the coding process, to let modelers do what they do best without compromising transparency and verifiability.

Initially, the model infrastructure was used as the underlying framework to develop 

, an analytics dashboard that displays historical and live data powered by financial risk models. 

It was met with great enthusiasm as users can visualize and immediately assess the health of lending protocols such as Aave & Compound; an example application that was built using our model infrastructure and our data ETL. It served as a primer as we built out our framework and continues to output new metrics as new models are built with the framework.


The Credmark Model Framework is a set of tools that enable developers to rapidly prototype and deploy models that use blockchain data via Credmark’s Ethereum data ETL. In the next few months, we will provide API access to this data.


The Model Framework is designed with modeling best practices in mind. We chose Python for the purpose of rapid model development and flexibility. Financial modelers familiar with the Web3.py library understand the difficulty of building models that need access to large amounts of blockchain data. Our toolkit eliminates much of the complexity thanks to:

intuitive interfaces to define models, including their input and output data so that they are readily composable,


model context to ensure deterministic behavior 


a model search engine so that users can find and use local and remote models,


common data objects such as Address, Contract, Token, Position, Portfolio that encapsulate data and smart behavior.

These are a few examples of the components that we’ve developed to streamline the process. Check out our 

When you fork our Framework repository in Github, you’ll find scripts and sample models to get you going very quickly. We’ve also provided basic algorithmic, protocol, and series models that will make your model building easier. For a complete list, look at the 

When the Framework moves into general availability, model authors will be rewarded for their work. Every time a model is built, the creator will receive an NFT which represents ownership of the model. The NFT holder can hold onto the NFT to receive a payout every time the model is used, or sell it off to a willing purchaser. 

Until the model NFT is ready, we suggest modelers include a wallet address in the bio section of GitHub so we can provide the NFT once the system is ready.

The Model Framework is in open beta. We have a long list of improvements and additions and expect that list to get longer as our community experiments and provides feedback. 

introduction of Membership NFTs which grants access to the financial modeling and analytics tools

 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.

An Introduction

Model Framework Open Beta Release

A few months ago I wrote about the difficulties of DeFi treasury management from a planning perspective. It was called “

DeFi Treasury Management - A Halloween Special

” because managing money in DeFi is a horror show. Normal DeFi risks, like losing all of your money when you type a single incorrect number in a 42 character Eth address, become really scary when you’re moving around tens (if not hundreds) of thousands of dollars. Add in a half dozen protocols, 3 chains, a series of incompatible tools and you can start to see why I had a horror show on my hands.

, who is responsible for Credmark’s treasury management strategy, had put together a very clear asset allocation plan. He’s an experienced asset manager and crypto trader. He knows what he’s doing.

You can find the plan for our first cycle 

It was published in mid-December. We planned to execute it by January 1st.


Asan’s treasury management strategy is relatively traditional, but on the aggressive side. We decided to keep the execution dead-simple:

We analyze the plan's performance at the end of the month.


Getting this done was my problem, so I was stuck executing. Because all of Credmark’s treasury is managed through multi-sigs, I also needed a volunteer. It was getting close to Christmas and everyone had something better to do, but my colleague 

 felt sorry for me and offered to help. It’s possible he’ll never forgive me.

We read through the plan together and said: “Two hours, max.” We scheduled a two and a half hour meeting, just in case.


We thought the first phase of our asset allocation was dead simple: diversify our stablecoin holdings. Our treasury was primarily USDC, with some USDT. Asan wanted a more diverse portfolio: more USDT, some TUSD, some DAI, and some PAX.

We store our treasury assets in a Gnosis safe, and we use CoinShift to make our lives easier. We executed our first swap on Gnosis: USDC for DAI. Collier set it up and I confirmed it.

20 minutes later, it failed. We tried again. Same result.
So we tried using CoinShift. Same problem, over and over.

Two and a half hours after starting, we still had 0 (ZERO!) DAI in our treasury. And that was only the first of many transactions.

At that point we called one of our in-house Ethereum experts and asked him what he thought was happening. After some digging, he reported that swaps from Gnosis (or tools built on Gnosis) requires the price for the swap to remain the same between multiple signers. If the price changes in between initiation and confirmation of the second signer, the transaction fails.

Even though the swap is initiated, the order put in may no longer be valid based on the new market price.

Further complicating things, if a swap isn’t executed within ~5 minutes of initiation, it fails. When you’re dealing with crazy gas fluctuations and coordinating across multiple continents, this time window is more constraining than you’d think.

Of course, all of this is configurable, but the usability isn’t even close to what we needed to get through our strategy during this decade. Collier and I settled on transferring the USDC to a Trezor, executing the swap, and then sending the DAI back to the treasury safe.

The next day I had to fly from Texas to Italy, so Collier and I couldn’t continue working together. We decided to clone a Trezor so that we could check each other. Not exactly multisig, but safer than putting all the responsibility on a single person.

When I got home, we scheduled our next meeting. Three hours, just to be safe.

If you’ve ever seen the movie “Home Alone” you probably remember the hapless robbers being foiled at every turn by the imaginative kid. Well, that’s exactly what my next few sessions with Collier felt like.

Every session we were sure we’d reach our goal. Session after session, we failed. In the end we spent 44 hours executing Asan’s plan.

Once we moved the USDC to our Trezors, swapping them for other stablecoins was easy. We used Uniswap and Sushiswap. We would have used 1inch, but Collier couldn’t get to the site because he was in the US. Thank you SEC.

But then Asan wanted us to buy some ADA, some SOL, some MATIC, and some BTC. 😱

One thing I forgot to mention earlier was that our core team members didn’t want us to use a central exchange. “Keep it DeFi” he said. We tried.

We had a bunch of ERC-20 tokens. How were we going to get our hands on tokens from other chains without an exchange? And once we had them, where were we going to put them? Even though Gnosis is on multiple chains, it isn’t really multichain yet. If you are on a non-Eth chain and try to send assets to your Eth Gnosis safe, POOF, you’ve just lit those assets on fire with no chance of recovery.

We started with SOL and came up with a cunning plan.

Use Wormhole Bridge to wrap ETH to move it to Solana.

Use a Solana DEX to Swap wrapped ETH for SOL.

Use Wormhole Bridge to wrap SOL to move it to Ethereum.

Unfortunately, everything that could go wrong went wrong.

Wormhole Bridge didn’t work with a Trezor connected to Metamask, so we had to move tokens to Metamask.

Solflare no longer allowed my Ledger to sign transactions. No idea why that stopped working. So I had to move SOL directly to a Solflare native wallet.

Later I had to destroy the Solflare native wallet to re-attach my ledger. No problem, I had the seed phrase. But when I recreated the wallet using the seed phrase, I got different addresses. An address that held USD 30K of SOL was inaccessible to me.

I spent two hours trying to come up with a story that explained how I’d lost thirty thousand dollars without making me sound like a complete idiot. My dog was puzzled by my pacing until I asked myself if maybe this was a wallet bug. Time for a Hail Mary 🙏🏻. I created a Solet wallet with the seed phrase and lo, it worked! The wallet contained the original address, and I could access our SOL.


As one of my colleagues likes to say: “We’re so early.”

. Very elegant. As I type, the service seems to have disappeared, so now I wonder if we’ll ever be able to unwrap our ADA! Sigh.

Next, Asan wanted us to lend some tokens and borrow others. He showed us a lot of math to justify all this activity, but Collier and I didn’t care by that point. We were the executioners, the brawn of the operation. Whatever the brain wanted, we were going to deliver, even if it killed us.

Luckily all of the borrowing and lending only involved ERC-20 tokens. That’s not quite true. Some BTC was required, but by that point we knew he really meant wBTC 😉

So we headed over to Aave and got to work. Incredibly, (almost) nothing went wrong. Everything worked as planned, at least on my computer. On Collier's, nothing worked. We don’t know why. Having a cloned Trezor saved the day.


We left a few token buys for last: QRDO and BLOK. We looked for them on every DEX we knew, but couldn’t find them. Time to break the rules.

We found a centralized exchange in Asia that allowed us to signup with only an email address. Lack of KYC made it feel almost DEXish. After a few tests we decided it wasn’t a scam, and bought some QRDO and BLOK. BLOK lives on the Polygon network. Using a centralized exchange insulated us from the pain of moving things on and off of Polygon.

The very last thing on our list was staking the BLOK. We found the app to do that, but it had taken so long for us to get through our treasury management plan we found that all the BLOK staking pools were closed. Our BLOK remains unstaked.


Our treasury management strategy has two very specific goals:

Earn a small profit on our assets without taking too much risk.

Guaranty enough liquidity to fund operations. 


At the end of this month our risk team will analyze how their strategy performed. We’re all looking forward to reading the results.

In the meantime we need to figure out how to make this process more efficient. Executing Asan’s strategy was not cheap. We spent about USD 3,000 in gas fees. As I mentioned earlier, Collier and I spent approximately 44 hours. Assuming our time is worth USD 100 per hour, that’s an additional cost of USD 4,400. Before we earn anything, we’re starting USD 7,400 in the hole.

Knowing what we now know, Collier and I believe that we could repeat this work in half the time. That’s still a lot. We will be talking to Asan about reducing the complexity of the strategy. Cross chain operations and exotic tokens were the most time-consuming aspects of our work. Eliminate those and everything will go much faster.

Sticking to DeFi was a great goal. Unfortunately, it’s not always possible. We could have saved some time and money if we’d acknowledged that from the start. Sometimes it’s OK to get a little dirty.

As I mentioned earlier, we planned to finish our asset allocation work by January 1st. We finished on the 23rd. By that time we should have been near the end of our first cycle. Market sentiment had changed dramatically. Asan’s second cycle plan would have taken that into account. Instead, he is going to have to deal with an additional month of mostly negative activity. Luckily, he’s a pretty conservative guy who understands the ins and outs of treasury management. The portfolio he had us build was carefully analyzed to minimize risk. That kind of planning is the difference between life and rekt.


The Execution

DeFi Treasury Management Part II

The analytics page provides deep insight to both the $CMK token as well as the staking of $xCMK. While the price statistics reveal understanding on trading volume and relevant markets, the staking statistics provide information on adoption of the $CMK token by the community.

As you need to stake $CMK to access our services, these statistics show effectively the number of users for our platform and how it develops over time.

The price statistics and market statistics provide insight on if a specific trading pair/exchange is favored by users or if volume/liquidity is evenly distributed. This helps for optimal capital allocation over trading venues as well as maintaining focus on the markets with the highest volume for capital efficiency.

With the release of the CMK Analytics page, we will display the current CMK Token distribution in a clear and interactive way. In order to do so, we assessed the way how we calculate the current circulating supply of the token and how this might need to change in order to be fully transparent about the current distribution of CMK.

Thus, we decided to divide the supply distribution into three parts. While Locked contains all vesting contracts, the 

 is made up from the DAO Treasury and the Community Rewards, while Public contains all public circulating supply. As the Treasury is not locked by a Smart Contract, we will count it in Total Circulating Supply, even though the funds are stored safely in a Gnosis Safe.

The implications of this change affect the

Total Circulating Supply, that is now roughly 61.5m CMK (up from 28.6m as shown before),

Market Cap, that is now roughly $16.1m (up from $7.5m as shown before), and

% CMK staked, that is now roughly 6.6% (down from 14.13% as shown before).

While this will artificially raise our reported market cap on third party sites, these reporting adjustments are designed to build transparency and trust, and don’t change the existing effective circulation of CMK.

As we scale up the platform and provide more risk metrics, we will see more CMK in the staking contract and more distribution on both CMK and xCMK. Throughout Q1 and Q2, you can expect our Risk Terminal to grow in terms of the number of protocols and models supported.

As the Terminal matures, CMK will continue to be staked as holders seek out key risk metrics. This effort will run in parallel with our partnership efforts as we establish clear problems within DeFi that our platform can address.

Distribution, or number of wallets staked and holding is a good metric for adoption. We also want to make sure that the treasury is well funded so that Credmark can weather any upcoming market events.

Once the platform is live in mid-2022, the partnerships we have formed will convert to API consumers and this will be reflected in our analytics page. It is a great place to track the success of our initiatives as we progress along our roadmap (this is coming soon!).

More to come! Follow along with us in our 

Risk Terminal Update: $CMK Analytics

My name is Peter and I recently joined Credmark as a Senior Data Engineer.

My undergraduate degree was in economics but I quickly pivoted into the tech space early on in my career. I got my Master’s degree in Data Science and it was through this program that I discovered Data Engineering. I was previously a Data Engineer at a startup where I created a Data Warehouse from scratch, built and maintained serverless data pipelines, and even dabbled in some DevOps work. Most recently, I was a Data Engineer consultant where I got the chance to build data pipelines for a lot of different companies from startups to major national e-commerce companies.

I first learned about Bitcoin in late 2016 from my mom of all people 😂. I started researching everything about Bitcoin and Ethereum and it became clear to me that this technology was going to radically change the world. Having smart contracts that could enable true peer-to-peer transactions in a decentralized, censorship resistant way was a very powerful idea. As I mentioned above, my undergraduate was in economics and my career has been in tech. Crypto is the convergence of those two disciplines so it was naturally a perfect fit for both my skill set and my personal interests.

As exciting as crypto was in 2017–2018 there was still no “killer app” that existed then to really showcase the power of crypto. It wasn’t until the DeFi boom in 2020 where I think crypto “killer apps” emerged that showed what could be created with smart contracts. I think DeFi has already shown that not only can it deliver on some of the promises of peer-to-peer transactions, but it can replace and outperform many of the same counterparts available in traditional finance all while being permissionless and open source for anyone all over the world to interact with.

I see DeFi as both a technical and financial revolution and I am excited to play a small role in helping to further the movement.

As a fellow DeFi degen, I find that the space is currently missing some basic risk metrics that are ubiquitous in the traditional finance world and that help investors gauge the level of risk associated with any investment. I love that Credmark is trying to bring these risk metrics to DeFi and doing so in an open and permissionless way and believe this data is a pivotal missing component for the space which will help attract more investors who may currently be turned off by the lack of clarity on the risks involved in DeFi.

My goal is to build a robust, open source data pipeline that can allow us to take a look at many DeFi protocols and compute risk metrics so that these can be incorporated into risk models and made available to all investors in DeFi. I want all investors to have a full picture so that they can properly weigh the risk and rewards of every investment decision in DeFi.

Success means, in the future, anyone will be able to help run and maintain data pipelines that deliver vital risk metric data for all major DeFi protocols.

Introducing Credmark’s Senior Data Engineer

From October through November of 2021, Credmark ran a hackathon looking for the best Uniswap v3 recommendation model on the market.

Even a cursory glance shows that uni v3 is not profitable without active rebalancing or a larger strategy

Uniswap v3 is seen as an undesirable product for retail since the barrier to entry, and position management overhead is quite high for a retail user. Although the capital efficiency is higher, determining the ranges in which to invest in a pool pair is another mountain to climb in itself. Having a recommendation model which suggests ranges and being able to see an estimated profit of the used strategy via simulation can give users a bit more confidence when providing liquidity to overcome the low rate of profit in many pools in Uniswap V3. The less volume, the lesser the fees for liquidity providers.

Taking a huge step in this direction we were using a Bollinger strategy for assessment in our Smart Pool and using our in-house Uniswap V3 Strategy Simulation Engine which gives an estimated APR for the strategy. This allows a user to understand the risks/profits for their assets in a more confident manner. But the APR numbers were not impressive enough.

Credmark’s Uniswap V3 Hackathon created a community of modelers for such cases which allows the developers to contribute to the mission of lowering the entry barrier to many complex protocols for a user.

And the community has proven us right. We are proud to announce that the winning model for our hackathon has outperformed our base Bollinger model. The level of sophistication shown by the winner in model development is impeccable. We are going to use this as our new base model in our smart pool application. This will allow us to highlight the work of our community and provide it a platform to be able to use retail users which can give them a huge advantage in making their decisions.

The top 7 model results can be found here:

And the winning model’s performance over a 4 week period of time can be found here: 

https://docs.google.com/spreadsheets/d/1sp5aPoHLnEQBReOLgGPqxYBHC8SYoQti7G2e5oOR5C4/edit?usp=sharing

More details regarding the competition can be found here: 

More importantly, below is the detailed technical document describing the strategy used by the winner. The following documentation is provided by the winner, demonsword, themself.

The purpose of the model mentioned below if to improve the accuracy of the prices of crypto currencies predicted by computers. In general, Bollinger is used for price prediction but it uses previous prices for a certain amount of time (for example 7 or 8 days). But LSTM as the name suggest remember changes for a very long period of time.

Long short-term memory (LSTM) networks were invented by Hochreiter and Schmidhuber in 1997 and set accuracy records in multiple applications domains.

Around 2007, LSTM started to revolutionize speech recognition, outperforming traditional models in certain speech applications. In 2009, a Connectionist Temporal Classification (CTC)-trained LSTM network was the first RNN to win pattern recognition contests when it won several competitions in connected handwriting recognition. In 2014, the Chinese company Baidu used CTC-trained RNNs to break the 2S09 Switchboard Hub5'00 speech recognition dataset benchmark without using any traditional speech processing methods.

The scope of LSTM is very huge as it can be used in Robot Control, Time Series Prediction, Speech recognition, Rhythm learning, etc.

But right now, we have fixed our scope on only time series prediction as the problem statement given to us is the perfect example of time series prediction. The model that I have made takes minimum six days data and can predict the data for next five days.

The only thing that we are predicting is the open and close price of crypto and then dividing it in three tiers. The model at least need data for six days to predict the prices for next five days not less than that.

For the prediction of prices, I have used three LSTMs. One LSTM is used to predict the max price of each day. Second one is used for minimum and the third one is used for median.

And on the basis of these three values for each day and the standard deviation I have calculated the prices for each and every tier.

First thing that I have done is to divide the dataset in three parts. One containing the maximum price of each day. Second containing the minimum price of each day. And the last containing that price of each day which are closer to median of that day.

Then I have to convert this time series problem into supervised problem in order to make it suitable for LSTM. So, I have used shifting for this process. And I have to shift the output of LSTM in order to get the prices of the future dates.

Now with the help of these prices I have designed an algorithm which will give ranges according to the tiers (High Risk, Medium Risk, Low Risk).

In general, there are eleven classical methods for solving time series.

Autoregressive Integrated Moving Average (ARIMA)

Seasonal Autoregressive Integrated Moving-Average (SARIMA)

Seasonal Autoregressive Integrated Moving-Average with Exogenous Regressors (SARIMAX)

Vector Autoregression Moving-Average (VARMA)

Vector Autoregression Moving-Average with Exogenous Regressors (VARMAX)

Holt Winter’s Exponential Smoothing (HWES)

For details about these methods you can check out this link:

https://machinelearningmastery.com/time-series-forecasting-methods-in-python-cheat-sheet/

But all these methods have one shortcoming that they cannot remember the trends which are way before in the past. And this problem can be solved by the use of LSTM. That’s why I have used LSTM for this problem.

First thing in any model is to prepare a dataset for your algorithm.

There are some columns in the dataset that have no need for the algorithm. Those columns are id, timestamp, timestamp1 and indexes. Next thing is to divide the dataset in three parts. For that I have collaborated all data entries according to datetime and from each group I only pick three entries first one is minimum, maximum and median of amount0 and add these entries in three different datasets.

After this we have to prepare the dataset for LSTM. This involves framing the dataset as a supervised learning problem and normalizing the input variables. We can use the shift() function in Pandas to automatically create new framings of time series problems given the desired length of input and output sequences.

This will help us to combine six rows combined to one row and converting this 11 columns dataset to 66 columns dataset. With these columns we have 5 more columns which will have the future token prices for next five days.

Now we can define and fit our LSTM model. We will define the LSTM with 50 neurons in the first hidden layer and 5 neurons in the output layer for predicting pollution. The input shape will be 6-time steps with 11 features. We will use the Mean Absolute Error (MAE) loss function and the efficient Adam version of stochastic gradient descent.

Now after this I have got the result in the normalized form. So, I have to un-normalize it. After unnormalizing it the work is still not complete. We will understand it with the help of an example. For example, we have the input of 10 days. So, input will be in the form of:

[ 1 2 3 4 5 6]
[ 2 3 4 5 6 7]
[ 3 4 5 6 7 8]
[ 4 5 6 7 8 9]
[ 5 6 7 8 9 10]

the numbers represent days in this matrix.

[ 7 8 9 10 11]
[ 8 9 10 11 12]
[ 9 10 11 12 13]
[10 11 12 13 14]
[11 12 13 14 15]

From this linear matrix I have to take only last five values as they are the only values associated with new days [11 12 13 14 15]. Now I have got minimum, maximum and median values for next five days.

Our next task is to create price ranges on the basis of three tiers. For that we have calculated standard deviation of all three datasets. For low-risk tier open range will be minimum predicted value and close range will be maximum predicted value. And standard deviation is added or subtracted to one of the open and close which is closer to median in order to increase the range between them. For medium risk tier open will be minimum range and close range will be maximum range. And for the high-risk tier standard deviation will be added or subtracted to the minimum or maximum which is farther from median in order to decrease the range. This will be the end of the model giving the prices of three tiers.

One of the main assumptions is that the current information has a dependency on previous data. This dependency is very important for the data.

LSTMs are sensitive to different random weight initializations

The input format will be the one given in hackathon. I have created a function named “convertinputformodel” which will take csv as input and return six things three matrices according to minimum, maximum and median and datetime for all three.

The output format of the model will be the prices of the next five days of the input . And then we have created price ranges of three tiers. As discussed in Model Methodology.

For Testing I have defined a term known as hit-ratio. Hit ratio is defined as the number of times price of token lies in the open-close range divided by total number of times tried.

 = price lies in open-close range / Total tries

As in many financial markets the price stays relatively stable and we can benefit from the multiple predictions provided by model for a particular day by taking average of them.

Uniswap v3 Hackathon Deep Dive

This past year has been the most rewarding, challenging, and exciting year of my life.

In January, after more than 2 years working on on-chain credit scoring as CTO, it was clear that Crypto was lacking the fundamental data and risk building blocks and I set out to fill those gaps.

In February, I took over as CEO of Credmark and decided that the future of risk was decentralized. I started to plan out what that would mean, and how to build a team.

In March, people caught wind of my plans, and before long, I was having conversations with investors. The appetite for DeFi risk data was substantial; VC’s were eager to invest in the project on the strength of Credmark’s previous work alone. I had earned the credibility through years of pavement pounding and finally had the resources to build a world class team.

I spent April onboarding (or in some cases, re-onboarding) my founding team and writing our 

. The Whitepaper was critical in clarifying my vision since Investors were eager to get in before we had a prototype. Drafting allowed me to think about the skillsets I would have to hire for to build the project.

, and got the wheels turning on releasing the token, CMK.

We wanted to get the token out quickly with the following assumptions:

A first-mover fair launch using Uniswap v3 would would grow a community of engaged experimenters to the project

The long bull market would turn around soon, and launching would become more difficult with time

Since we had runway, Credmark owning it’s liquidity was more important than cash

With that in mind, we set out planning the launch. Uniswap v3 was about to launch and seemed like a perfect candidate to use for an 

. I studied the documentation Uniswap released on the newest iteration of their protocol and saw immense opportunity to fair-launch on the platform. We would be the first token ever to launch on Uniswap v3, an experiment in creating a fairer and simpler alternative to other IDO platforms.

In June, we launched the token. My desire to experiment and innovate created two problems: many people didn’t understand the mechanism, and thought that we were rug-pulling, rather than fair-launching (Since then Alchemist’s Copper has done similar launches with far better results). In the confusion, we lost some of the support of the community. Adding to our list of complications, the fear we tried to avoid hit; the 3 days of our token launch were some of the worst in the crypto markets in 2021. As a result, we weren’t able to rely on market confidence in order to help us with our goals.

While a less-than-flawless token launch was inconvenient, Credmark was founded to solve risk and data problems in DeFi. We needed to put our heads down and build: So we did.

In July, we used our existing knowledge of Uniswap v3 to 

. It was our Proof of Concept for how an application could use Credmark’s data in order to perform a DeFi function, and give retail DeFi users a tool that previously would only be available to institutions. The problem with concentrated liquidity is that without enormous amounts of capital (i.e. being a Whale), it has abysmal returns.

Our experience with SmartPool allowed us to refine our roadmap. We knew who we needed to hire, and go out into the market to better understand risk-management problems folks in DeFi were having.

August & September we continued to grow our team and set up relationships and partnerships. Our partnership strategy was to find DeFi projects that have risk exposure and determine whether tools like SmartPool, or the data driving it, could help. We established research, service, and data partnerships with 

In September we also finished up the base layer of our Data ETL. Originally our plan had been to buy, rather than build, a data solution. However, there was nothing available on the market sufficient to fuel a decentralized modeling platform. Clearly it was something that had to be moved in-house. While it extended our build timeline, I am very excited to be able to offer the type of quality data-as-a-service to other projects.

To prepare for our platform launch, we also established relationships with Market Makers and Centralized Exchanges. These relationships will drastically decrease the barrier to entry for acquiring CMK, a necessary component to interact with our platform.

October was spent building out our Platform. We decided to do the platform in 2 phases, the first is single sided staking with access to graphs of some of our Risk Primitives, VAR and LCR. The second is to add revenue sharing to the staking mechanism via our Data Platform Roles and a self-service API. We also released our 

 in October, a repository for our research and information about the DeFi landscape.

November saw the expansion of the risk library and the onboarding of the final members of our leadership team. It was also the opportunity for us to finish and test the first phase of the platform.

December was Credmark’s first On-Site in Austin TX, with Asan, our head of Risk and Research, Matthias, our head of Product, and Paul, our COO coming to town to spend time with myself, Emilie, Momin, Stijn, Joey, and Collier. We spent a week finalizing process and organizational structure as we grow from a small to a medium sized startup.

 at hand within the first 2 weeks, we’ve had over half a million dollars staked in order to access our risk metrics!

Here’s to growth in 2022 and making DeFi a safer, more deeply understood place!

Credmark 2021: A Retrospective

) as well as some insights into how we are building our own 

, we explored the outputs of the Credmark platform, with focus on the first 2 models we produced: 

This post will expand on the following questions:

How can these metrics be used in a composable manner with existing metrics (like TVL)?

Liquidity risk is the risk that a business will have insufficient funds to meet its financial commitments in a timely manner.

To measure this risk, we use a metric called liquidity coverage ratio (LCR). LCR is defined as the proportion of highly liquid assets held by an organization to ensure that they maintain an ongoing ability to meet their short-term obligations (cash outflows for 30 days) in a stress situation. It is one of the essential liquidity risk measures in traditional finance.

Value-at-Risk (VaR) is a model to measure market risk. VaR is a statistical measure used to calculate the maximum potential loss caused by market movement under business-as-usual circumstances.

It is defined as the maximum dollar amount expected to be lost over a given time horizon, at a pre-defined confidence level, given normal market conditions.

It’s important to note that this measure doesn’t provide any guidance on how much can be lost in a “black swan” type event, an unknown event with unknown probability.

Credmark Blog

For the latest technical analysis, partnerships, and community updates