mirror of
https://github.com/luxfi/consensus.git
synced 2026-09-24 00:26:58 +00:00
zeekayandHanzo Dev
4abf37c952
The live guard refused to start a chain whose K was below the number of validators that existed. But sampling is stake-weighted: a validator is drawn in proportion to what it staked, however many validators there are, and K only sets how many draws a poll makes. Tying K to the headcount gave a validator holding a billionth of the stake the same standing as one holding a fifth, and refused to boot a producing chain because a small staker had joined. The guard keeps what makes a committee safe — the overlap bound and at least single-fault tolerance — and stops counting heads. The tier floor stays a reported target: refusing to start does not add validators, it pins every node on whatever it last booted. Co-authored-by: Hanzo Dev <[email protected]>
19 lines
733 B
Go
19 lines
733 B
Go
package config
|
|
|
|
import "math"
|
|
|
|
// ConsensusSuperMajority is the 69% agreement threshold (LP-CONSENSUS-69),
|
|
// two points above the traditional 67% BFT bound.
|
|
//
|
|
// This is the float form, for sizing a sample. Deciding whether a given set of
|
|
// votes carries a quorum is exact integer arithmetic over stake — see
|
|
// quorum_threshold.go. Do not reintroduce a float predicate here: rounding
|
|
// twice, two different ways, is how a chain disagrees with itself.
|
|
const ConsensusSuperMajority = 0.69
|
|
|
|
// AlphaForK returns the votes needed out of a K-node sample to reach
|
|
// ConsensusSuperMajority, rounding up so the threshold is never understated.
|
|
func AlphaForK(k int) int {
|
|
return int(math.Ceil(float64(k) * ConsensusSuperMajority))
|
|
}
|