I built a Layer-1 as a reference implementation, and in the process hand-rolled some cryptography I'd value this list's review of the exact kind of thing that hides subtle bugs.
The two constructions I'd most like eyes on:
ECVRF over edwards25519 (RFC 9381-style), where I reuse each participant's Ed25519 signing key as the VRF key to seed an un-grindable randomness beacon for leader election. I validate the proof's gamma is in the prime-order subgroup and reject low-order points. My concern is the key-reuse composition (same key for Ed25519 signatures and as a VRF key) and whether my instantiation deviates from RFC 9381 in any way that weakens unpredictability or uniqueness.
A Sybil-resistance argument for concave stake weighting. Block-building weight scales with √(stake) to blunt wealth concentration, but concavity is splittable (k·√(S/k) = √k·√S), so I gate weight by an on-chain personhood score from independent attesters. I claim this reduces Sybil resistance to "the cost of obtaining independent attestations" and I'm explicit that it does not solve proof-of-personhood. I'd like the argument stress-tested.
It's MIT-licensed, ~2k lines of Python, with a test suite, a property-based fuzz harness, and a written security review that already documents known limitations (e.g. the reference VRF is not constant-time). It's experimental and secures nothing of value, I'm looking for scrutiny, not users.
Code and whitepaper:
https://github.com/gendpaul/cairnThanks for any time you can spare tearing at it.
David Paul
