Loops HouseResidual

NEAR HACKATHON
Links
Team
1 member- AROwner
Arifa Khan
Overview
Residual is an open-source, read-only NEAR post-quantum access-key migration checker. NEP-645 enables ML-DSA-65 keys alongside existing Ed25519 and secp256k1 keys. An account can therefore show post-quantum key adoption while retaining an independently usable classical full-access path. Residual reads access-key lists at a pinned finalized NEAR block and distinguishes five observed states: classical-only, mixed post-quantum/classical full access, PQ-only full access, PQ function-call-only and unresolved. It keeps missing/failed reads in the denominator rather than counting them as safe. The browser lets judges check a public account and sample recent signers, while the Python CLI emits reproducible JSON/HTML reports. The classifier has a Lean 4 model and 128-case Python/browser parity checks. This measures the observed access-key layer, not end-to-end quantum safety. Demo: https://agentzeta.github.io/residual-near/ . Source: https://github.com/agentzeta/residual-near .
The Problem: NEAR's introduction of ML-DSA-65 post-quantum signatures creates a critical measurement problem: adding a quantum-resistant key does not automatically retire existing classical full-access keys.
An account may therefore appear to have adopted post-quantum cryptography while still allowing transactions authorized by Ed25519 or secp256k1 keys.
We need a reliable way to distinguish post-quantum key adoption from the actual retirement of classical full-access authority.
Residual addresses this gap using observable onchain evidence, while explicitly preserving uncertainty.
We built a live browser application, Python command-line scanner, deterministic sampling engine, JSON evidence reports, and a static dashboard. The system records block provenance, distinguishes resolved from unresolved observations, and supports reproducible analysis.
We also developed a Lean 4 formal model and a 128-case truth table to test classification behavior.
A sample at block 224577581 resolved 5 accounts.