Loops HouseNEAR, Cross-Examined

NEAR HACKATHON
Links
Team
1 member- ALOwner
Alan Finney
Overview
Problem: crypto theses run on conviction. NEAR's usage has fallen sharply while its token is priced for growth, and a reader cannot tell which claims the data supports.
Product: NEAR, Cross-Examined is an investment thesis on NEAR that four AI models from four makers (gpt-oss-120b, DeepSeek-V4.1-Flash, GLM-5.2, Qwen3-30B) argued claim by claim: 20 claims, bull and bear, one a planted false claim as a control. Each model judges every claim separately and cites only metric paths; code fills in the numbers.
Thesis: NEAR is a one-product network priced for that product to keep compounding. Users paid NEAR apps $59.3 million in fees in 365 days, about 51 times the chain's gas fees (a different flow: scale only), and 87% of the last 30 days' app fees came from NEAR Intents. Active accounts are down 78% and transactions 85% (30-day averages vs a year ago).
Track question: does NEAR Intents activity lead network usage better than price? Weekly changes, led 0 to 12 weeks, over 79 weeks: neither leads. The largest absolute correlation at 4 to 12 weeks is 0.02 (Intents fees) and 0.08 (price); every 95% interval includes zero.
Check it: make verify needs no key or network; in a fresh clone it checks the 954 derived numbers against recorded hashes, rebuilds the BigQuery numbers and checks 103 citations; make verify-online re-downloads every source and rebuilds all 954. NearBlocks, DefiLlama, CoinGecko and Intents Explorer files are hashed, not shipped (their terms). BigQuery matches NearBlocks on all 365 days; transactions fell 72% in one day (2 Apr 2026), cause unknown. WEAKNESSES.md lists ten weak points.
Falsifiable: six tripwires and three forecasts, due 2027-02-11, scored by score.py. If two of T1, T2 and T3 are hit, the thesis is wrong.
How NEAR fits: all data is NEAR chain or Intents data.
Built by 3Dogs Nexus (3dogs.ai). Built on 3Dogs wisdom of crowds, three reviewers for speed, five when one stands alone. The code was written with AI coding tools.