Inspiration

Testnets drop test eth and it's nice to send it back when you're done playing with it. Developers need test eth to build things. Testnets have faucets that drop these test eth when requested. These need to be mined, and there are very few miners as there is little economic incentive to mine test eth. This means that sybil attacks are easy because it's easier to get consensus. Also, since transaction costs are zero, long-running spam attacks are possible if one has a lot of test eth.

This lead to some killjoys abusing testnets. They were hoarding test eth and then performed spamming or sybil attacks. To prevent this, testnets require Proof of Work and mining. This also means that they tend to be low on eth, as there are few miners.

The easiest way to help a fellow dev is do donate back the test eth to the testnet. This will save the dev the work needed for PoW.

Now it's turn for us to give back to these donors who have saved us all the PoW work.

What it does & How we built it

We first comb through mountains of data on Ethereum testnets to identify which wallets donate back to testnet faucets. We do a lot of analysis here, such as removing filtering out addresses that are smart contracts or that do not have any activity on Ethereum mainnet.

Once the analysis is finally done, we encode those addresses into a highly compressed data structure called a merkle tree, and deploy a modern ERC721 using the merkle tree as a fancy whitelist. It’s worth noting that any given address can only mint once, and that these mints are free (excluding gas costs). This is not for profit.

The NFTs in the smart contract are AI generated images, using a state of the art diffusion model to create them. This is computationally quite expensive, but the end result is very satisfying. These images are then stored ipfs and pinned with Filecoin, to ensure they’re never lost. They are, truly, decentralized.

The last part of the equation is the frontend, the website where all comes together.

We let top donors mint cool NFT art with water droplets. To do that, waterdrop does three things:

Whitelisting on-chain
- Gets the list of top 1000 donors to generate a whitelist
  - analysis of gazillion blocks
  - exclude smart contracts
- (Bonus) publishes leaderboard on our website that anyone can view (flask, tailwind, jinja2)
- Creates a merkle tree where the rootnote allows to check for eligibility for the mint
Generating the NFT art and publishing it to IPFS
- Runs an AI model to generate thousands of images from text snippets about water drops (Dalle Mega)
Minting the NFTs
- Creates a frontend where unique waterdrop NFT art can be minted by whitelisted wallets
  - Deploying a smart contract that checks wallet against whitelist

Challenges we ran into

The largest challenges were the gas cost and the blockchain data analysis. Initially, we thought about dropping NFTs to top donors, but gas was prohibitively expensive, so we understood we had three options:

deploy on L2
deploy on Polygon
merkle trees and letting donors mint their NFTs instead of dropping them

We decided to go with the 2nd and 3rd option, because it was more interactive. To do so, we had to create a smart contract with a whitelist. Our whitelist has 108,367 bytes, but using merkle trees we were able check wallets against a 64 byte rootnode. This requires some computation but at least doesn’t cost hundreds of thousands of dollars in gas. In fact, a mint at 10 gwei gas prices is less than $3! Then we analysed Polygon and deployed on Polygon.

The blockchain data analysis was difficult because of the ~15GB of data on the testnets and because most of it are not donations. On top of that, most interactions are smart contracts, not wallets, so these had to be filtered out as well. And lastly, we wanted to filter for wallets that also were active on L1, so that was an added complexity.

Generating interesting art also turned out to be more challenging than expected. Konrad started off with mathematical, procedurally generated art. The idea was to use variables such as donation size to determine the look and complexity of the artwork. However, it quickly turned out that it the artwork tended to look relatively similar. That's why we decided to use our expertise in AI to leverage it to create an AI-generated NFT collection, because there is much greater variety there. However, it took many many hours of computation.

Smaller challenges:

Getting comprehensive list of faucets required a surprising amount of manual digging
Integrating the entire pipeline (Solidity, Backend, Frontend, AI & Data) was not at all trivial
In particular, dealing Solidity versioning and merkle tree proof formats was a PITA
smaller frontend issues

Accomplishments that we're proud of

We’re very proud that we can give back to our wonderful community. We find rewarding merit through retroactive public goods to be a very attractive concept. Testnets are a great way of testing whether someone is a good or bad actor, because they provide no external incentive. You stand nothing to gain or lose from donating or hoarding or attacking a testnet. Rewarding those who donate their test eth back will not tip the scales, but it’s a nice gesture, just like their donation was a nice gesture.

We are also proud of the pace of progress. Hugo ran the entire backend ops on his own, and also created the entire gallery of art. Konrad had zero programming skills and understands the basics of frontend now and expanded his stack stack by VSCode, git, tailwind, flask, jinja in just 3 days.

What we learned

Hugo learnt:

Applying data analysis skills to on-chain data
The beauty of Merkle trees
Optimizing gas costs
AI image generation using diffuse models
Secure, decentralized permastorage on IPFS/Filecoin
how to survive a trip through frontend hel

Konrad learnt everything from scratch. His previous experience was only some basic html and CSS ages ago. He learnt:

using VSCode
the basics of
- git
- python
- flask
- tailwind
- Perlin noise, procedurally generated art

Together, we learnt about Kanbans, project management, collaborating using git.

What's next for Waterdrop

Hopefully we inspire other projects to care about the public good, and incentivize more good actors in the space to appear.

Regarding waterdrop, the whitelist to mint is updateable. Taking another snapshot at a future point in time, or even at regular intervals, is an option. Of course, the fundamentals behind waterdrop can also be applied to many other topics, not just testnet ether donations.