Under Wattenhofer & Kniep's New Constellation (v1.0), each 0.8 s window belongs to one hub and has k concurrent proposers sampled from that hub's validators. Instead of threshold-splitting a transaction across t-of-n proposers (collusion explorer), this sender sends the whole transaction to one proposer it trusts, and only in a window where such a proposer is in the set. How many windows can it use, and how long does it wait? Validator snapshot: Nov 20 2026, 9:49 AM ET. New here? Read the explainer.
JitoLabs+GBX / Firedancer+GBX and are
not in the "Independent" or "Vanilla" presets.
Trusting a scheduler type means trusting every operator running that client. Clients are self-reported clientIds from gossip.getVoteAccounts stake, getClusterNodes
clientId, ipinfo.io geolocation of the gossip IP (approximate), and on-chain validator-info names where published.
GBX has no gossip clientId, so its 14 connected validators (3.2% of stake) come from Titan's keyless
GetConnectedLeaders (Nov 20 2026, 11:45 AM ET), every slot cross-checked against getLeaderSchedule.
Validators without geolocation (≈6% of stake) never propose in hub mode. Hubs with fewer than k validators or less than the
min hub stake get no windows, and their validators are not moved. Distribution overrides rescale stake inside one hub only.GetConnectedLeaders)