Loops HouseProof of Numbers

NEAR HACKATHON
Links
Team
1 member- NOOwner
noegng
Overview
Problem: dozens of teams in the thesis track measured NEAR independently and published different numbers for the same quantities. A reader cannot tell which differences are errors and which are definitions, and matching numbers can pass for independent confirmation.
What we built: a replication study. We recorded 367 published numbers from 59 thesis repositories, each with its verbatim quote and the commit that wrote it, and recomputed every one from raw public data (DefiLlama, NearBlocks, a public archival RPC). Compared on the day each was published, 317 are within 5% of our recomputation. Where teams differ, the cause is mostly the definition: window length, net or gross issuance, gross fees or captured revenue.
Two findings no single thesis shows. First, every Intents volume and fee number traces back to one upstream dataset, so agreement between teams is one source read many times. Second, captured revenue can be checked independently: it arrives as NEAR in three wallets, and their growth read from an archival node is 0.98x to 1.01x of reported revenue over 30, 90 and 360 days.
The thesis built only on numbers that replicate: base-layer activity fell about 85% in a year, while NEAR Intents revenue now offsets 13% to 19% of net issuance depending on the window. Owning NEAR is therefore a bet that captured revenue grows about 5.4x, not a claim that it is cheap. Five kill conditions and four dated predictions run as code: python3 -m replicated.checks fails if the thesis should be dropped.
How NEAR fits in: the evidence is NEAR on-chain data: block headers (total supply), balances of the Intents revenue wallets at 15 past block heights, the validator set and the protocol config.
Reproducible: Python standard library only, raw data committed with SHA-256, 44 offline tests (one fails if a figure in THESIS.md differs from the computed metric), and a daily job that re-tests the thesis on new data. Live page: https://noegng.github.io/proof-of-numbers/