Loops HouseNEARESIS

NEAR HACKATHON
Links
Team
1 member- AUOwner
Austin Jeremiah J
Overview
Most crypto research takes three different things, what was actually observed on-chain, what someone calculated from it, and what they concluded, and squashes them into one confident chart. You usually can’t tell which part is data and which part is opinion, and you can’t check any of it yourself.
We built Nearesis to keep those apart. You ask one question about NEAR, and an agent takes it all the way to a published thesis. Before it looks at any data, it writes down what it’s testing and what result would prove it wrong, and locks that plan. Then it pulls real data from NEAR’s public BigQuery dataset, the NEAR Intents Explorer, 1Click quotes and CoinGecko. It writes a thesis that shows the evidence for the claim and the evidence against it, side by side.
The agents never write SQL. They can only ask for metrics we’ve defined and tested, and every number they get back has a hash, so you can trace it to the exact file it came from. Missing days stay missing, we never fill them with zeros.
NEAR is part of how the results get checked, not just the topic. Each result is frozen and re-run inside an OutLayer TEE on NEAR testnet, so anyone can look up the transaction and confirm the calculation. A full run costs about one cent.
Our own thesis, Volume Without Value Capture, looks at NEAR Intents. For every $100 settled, users pay about 59 cents, but NEAR keeps less than a tenth of a cent as gas. We also say where we could be wrong. Our forecasts are weak, and part of the flow is now private, so we can’t see it.
To try it, open the live app, or clone the repo and run pnpm thesis:build and pnpm thesis:verify. That rebuilds the whole thesis offline and checks every hash.