==Vibe Coding Tutorial: On Chain Data Research: Aztec Crowdfunding== Coding Agent is not only a programming assistant, but also a powerful tool for data analysis. With the free data provided by Etherscan, we can easily start data analysis at a low cost. The recent CCA crowdfunding campaign conducted by Aztec on Uniswap is an excellent research case to see to what extent programming agents can achieve. one ️⃣ Download on chain data There are many ways to obtain data on the chain, but Etherscan provides some free quotas that can be used. Find Aztec's crowdfunding contract (you can ask LLM to help you find it, or go to the official website to find it), and then download the CSV of all transactions on Etherscan. >Contract address: 0x608c4e792c65f5527B3f70715dea44d3b302f4ee, CSV download can be found in the CSV Export section at the bottom right corner of the page Here's a key trick: Since Etherscan has a maximum limit of 5000 CSV exports per transaction, you need to adopt a "saturated export" strategy that covers every day (even if there is time overlap). If the trading volume exceeds 5000 on a certain day, there must be omissions in a single export, and this method is not applicable. API or other methods need to be used instead. You need to export multiple times in batches to ensure coverage of all data within the cycle, and you will eventually obtain approximately 20000 records. Just throw this bunch of CSV files into a separate folder. Don't worry, let AI handle everything in the future. two ️⃣ Try using Codex I heard that OpenAI's GPT-5.1-Codes-Max is very strong. Let's test it out this time. Open the terminal and enter the folder where the CSV file was stored earlier. (Don't know the instructions? Ask AI.) After installing the Codex environment, enter the interactive interface. (Don't know the instructions? Ask AI.) After logging in, you can start analyzing. Of course, you can also use Trae/cursor/Claude Code instead, the logic is the same. three ️⃣ Merge documents Due to the potential data redundancy between different CSV files caused by batch downloading, we need to clean the data first. You can directly send instructions: >I have uploaded a bunch of transaction CSV files for the Aztec protocol crowdfunding process, but there are duplicate data between the files. Please help me merge these CSV files into a clean, non duplicated full CSV file, and we will conduct research based on this file in the future. It will automatically help you merge. If you are not confident, you can have it report the total number of merged rows and compare it with the total number displayed on Etherscan. This can usually be achieved in one step without any errors. If you spot check and find an error, then tell it specifically which one is wrong, and it will help you locate and fix it. Tip: Don't just give mechanical instructions, it's best to tell the AI the context of the data (which is crowdfunding data) and your subsequent plans, so that it can better understand the meaning of the fields. four ️⃣ Confirm key details One key detail of Aztec's mission is that 25% of the quota is exclusive to early users, and after a certain point in time, everyone will compete fairly. By searching (or AI), it can be found that the key point is December 2, 2025 at 14:00 UTC. You can have AI confirm: >I have found the official documentation, and the entire Aztec crowdfunding is divided into two parts, the first part is limited and the second part is public. The public bid starts from 12-02 14:00 UTC, please help me confirm this with the data. Its result is: >(Ignoring a few details) Conclusion: It is reasonable to use 12-02 14:00 UTC as the public open boundary. The data shows a significant increase in volume and single transaction amount during this period, which can support the official statement. five ️⃣ How to study? If you have a clear research direction, such as average amount, total fundraising amount, proportion of large investors, etc., just ask the Coding Agent directly. If you don't have a clue at the moment, why not try this "open-ended question": >Please fully explore the data inside and find at least 10 interesting and counterintuitive Insights. Its result is: =========== -Extreme concentration in the head: 65.9% of the total amount comes from the top 1% (232 transactions), 95.1% comes from the top 10% (2316 transactions); Long tail with a large amount of small amounts. -The proportion of small/zero transactions is huge: 44.2% of transactions are less than 0.01 ETH, and 7.85% are completely 0 ETH, which looks like a test or placeholder bid. -Large amounts dominate the total amount: only 450 transactions (~1.9%) with ≥ 10 ETH contributed 77.4% of the total amount; Among them, 104 transactions are ≥ 50 ETH, and 62 transactions are ≥ 100 ETH. -Single transaction ceiling: maximum single transaction of 250 ETH (2022-12-01 14:35:35, Submit Bid); Almost all 62 transactions of ≥ 100 ETH occurred after the open window. -Single hour ceiling: Received 1789.56 ETH from 14:00 to 15:00 on December 1, 2025, the highest throughout the entire process; The second highest is 12-01 13:00 (1420 ETH), and the third is 12-06 02:00 (1197.54 ETH). -Cut point effect (12-02 14:00 UTC in your document): The average value of a single transaction before and after the cut point ranges from 0.92 ETH to 1.92 ETH; The proportion of the amount increased from 45.2% to 54.8%; 14: The hourly peak of 911/613/559 ETH occurred in the last 3 hours after midnight. -Extremely low and discrete fees: 95th percentile handling fee is only~0.000617 ETH; The highest fee of 0.00622 ETH appears in a 0.40 ETH Submit Bid (rate~1.5%), which is an abnormally high fee small transaction. -Full hour transactions with zero amount: 2025-12-06 18/19/20:00, 2025-12-07 03:00. These hours have transactions with zero total amount, which is basically a zero amount operation triggered by non Submit methods (such as Sweet/Claim/Exit). -Method 'Asymmetric': All 1818+transactions (Exit/Checkpoint/Claim/Sweep) except for Submit Bid have Value-IN=0, indicating that the file records management actions but does not have corresponding outflow fields (Value_SUT is also all 0). The fund side is a one-way entry. -Participant concentration: The highest investment address is 0x4c8c... 14m3, accumulating~288.28 ETH (16 transactions); The highest number of addresses 0xe88c... 79e8 have submitted 27 transactions, but the cumulative amount is still~212.22 ETH, indicating that high-frequency addresses are not purely "small brush volume". =========== Based on these clues, you can continue to delve deeper through the new dialogue window. six ️⃣ AI may not be reliable, remember to verify If it's just personal exploration, the conclusions of AI can serve as a good starting reference. But please remember, AI is just an assistant, not the ultimate referee. I did not manually verify each of the 10 conclusions summarized by AI above. If you want to use this data for official publication or investment decision-making, please be sure to personally verify the accuracy of the data. Treat AI as an amplifier of inspiration, rather than a filter of facts. seven ️⃣ Conclusion The essence of Vibe Coding is not to write code, but to empower individuals with stronger exploration and analysis abilities. I hope this practical experience can show you that data mining is no longer the privilege of professionals, but a tool within everyone's reach.