Loops HousePayband

NEAR HACKATHON
Links
Team
1 member- ADOwner
Aditya Kumar
Overview
Pay transparency helps the people who are underpaid most, and they are the least willing to share their number. Payband lets a group learn its pay bands without anyone revealing a salary.
Each member's browser fetches a fresh NEAR AI Cloud attestation, then encrypts {role, salary} to the attested key of a model running in a TEE (NEAR AI's end-to-end encryption). Only ciphertext is stored on a NEAR contract. After the deadline, every ciphertext goes to the model in one sealed request; it returns only the 25th percentile, median and 75th percentile per role, and only for groups of at least four. The contract's publish method parses the signed request and refuses it unless it is exactly the instruction pinned when the round opened plus each stored submission once, as its own message, in order; it verifies the model's ed25519 signature over the exact request and reply on chain, and enforces k-anonymity per group. The reply is encrypted to a throwaway key that is published, so anyone can decrypt it and verify the bands in the browser.
How NEAR fits: NEAR AI Cloud provides the attested, end-to-end encrypted inference; the NEAR contract verifies the enclave instead of trusting whoever runs the aggregation; function-call access keys let members submit with a key that can only call submit.
Working on testnet with real NEAR AI Cloud inference: 19 contract tests (including injected, duplicated, reordered and dropped submissions), scripts for a full round, a live page with in-browser verification. The model does the arithmetic; the README compares its bands with exact percentiles (5 of 6 exact, one quartile $2k low). A deterministic aggregator for OutLayer's TEE is written and unit-tested but not live: OutLayer testnet runs were timing out for all teams. Same author's other entry: Sureplate.