The question
Which post-quantum ZK proof system should the Mantle proof-of-leadership (PoL) circuit use (and then other Logos blockchain circuits)? Constraint that dominates the choice: proof size (every node verifies every PoL proof), with ZK required (the proof must not link to the leader's identity or stake).
The answer
Goldilocks STARK over an RPO-based AIR. It is transparent, fully implemented on a tested prover, and the smallest option that actually works today: ~45 KB non-ZK, ~47 KB with ZK, ~41 ms prove, ~0.87 ms verify. The other candidates lose primarily on proof size.
Why the alternatives lost
- Lattice (LaBRADOR, RoKoko, RoK-and-Roll, SALSAA): the more compact long-term bet, but not yet usable. LaBRADOR's ~50 KB exists only in AVX-512 reference code; the portable CPU build measures ~2 MB / ~30 s and its verifier is linear. The others ship only estimators or partial harnesses. Revisit as implementations mature.
- Vortex (lattice+hash, in production on Linea): fastest prover, but on a realistic PoL relation the proof blew past 200 KB once the matrix wasn't perfectly square and ZK hiding was added. The blow-up is specific to the linear-code (Ligero/Vortex) O(√N) family and does not affect FRI-based systems.
- 31-bit fields (BabyBear/KoalaBear/Mersenne31), the industry direction: optimize prover throughput, not proof size. The same PoL circuit produces 0.9–1.8 MB over Plonky3's 31-bit paths, i.e. 20–40× larger for the identical statement.
Two field/hash choices worth flagging
- Goldilocks (64-bit), not a 31-bit field: the ticket and threshold are the size of a hash-function output element and don't need 64 bits, but the note value
v does. Keeping v in a single Goldilocks element avoids multi-limb arithmetic with carries and range checks in the hot path, which is where a 31-bit field would spend its small-field speedup re-implementing 64-bit math.
- RPO, not Poseidon2: RPO is AIR-native (7 rounds → ~4× shorter trace than Poseidon2 here). Poseidon2 remains the right choice for the R1CS/SNARK side; a Poseidon2-over-Goldilocks variant is a future option if hash-family fidelity with the SNARK spec matters more than prover speed/size.
Open items
- Benchmark table was AI-generated on a modulus-reduced circuit: we needs independent verification.
- ZK numbers are proof-size only; the ZK guarantee rests on a simulator argument the functional soundness tests do not exercise.
References