Loops HouseKEYTELL

NEAR HACKATHON
Links
Team
1 member- KIOwner
Kingsley Ojilere
Overview
Keytell tests one investment thesis about NEAR and publishes the result either way: are transactions signed by function-call keys (the scoped keys apps and agents hold) a better leading indicator of real usage, measured as weekly USDC volume, than daily active accounts or price? Only NEAR has protocol-level function-call keys, and the kind of key behind every signature is public.
The claim, its pass rule and the SHA-256 of the method and every query were written to a keyless NEAR contract with no edit method before any out-of-sample data was read. Registered 2026-11-13, before the first out-of-sample query.
Result: The claim did not hold. Over 94 weeks the test had never seen, adding function-call activity to a baseline forecast of weekly USDC volume made it 5.1% worse; daily active accounts made it 3.6% worse, price 5.5% worse. It needed to beat both. It beat price but not daily active accounts, and neither difference was clear of noise (both bootstrap intervals straddle zero).
What we found instead: signing moved from app-held function-call keys to relayed meta transactions (NEP-366). 1 of 8 control reruns would pass the registered rule: every relayed meta transaction counted as function-call activity. It was registered as a control, not as the claim, so it does not change the verdict; docs/FINDINGS.md treats it as a hypothesis for a new pre-registration.
How: a keyless sampler classifies the key that signed 50,600 mainnet transactions at signing time over public RPC; BigQuery counts DAA and USDC; the app shows every chart's SQL, the case against (controls, 1,000 timing placebos, concentration), a /keys lookup for any account, and a memo the analyst signs with NEP-413. A twelve-week forward claim is open on chain; a crank posts each week with a function-call key limited to post, and anyone can resolve it. make verify rebuilds every number from a clean clone. Ledger on testnet; data from mainnet.