mirror of
https://github.com/luxfi/consensus.git
synced 2026-09-24 00:26:58 +00:00
zeekay
07ffec05a7
CR-12 — bind effectivePolicy byte into ComputeRoundDigest transcript protocol/quasar/round_digest.go: new uint8 parameter, refused if zero, bound as a TupleHash part between verifierBytes and netBytes. Closes cross-policy replay: an adversary can no longer sign a Round R as PolicyPQ and strip the Q witness on retransmit, presenting the bare P witness as a PolicyQuorum cert under an identical digest. Tests updated: canonical inputs include effectivePolicy=1, FixedHash- Family mirror parts include the policy byte at the matching position. CR-13 — rejection sampling closes modulo bias in three samplers protocol/photon/emitter.go cryptoRandInt protocol/prism/cut.go cryptoRandInt protocol/prism/stake_weighted_cut.go cryptoRandUint64 Prior `% max` form was structurally biased for non-power-of-2 max; under nation-state grinding model (Pinkas-Reiter style) the bias was an exploit on committee sampling. Rejection sampling computes the largest multiple of max that fits in uint64 and rejects values above it. Expected iterations ~1-2; never more than 2 in expectation. Tests pass on protocol/quasar, protocol/photon, protocol/prism. BLOCKERS.md CR-12 and CR-13 marked closed. Two of 13 critical findings down. Remaining 11 are in progress.