13.85% of stake across 93 of 779 validators is enough to make Agave/Jito monoculture less embarrassing, but not enough to let a C/C++ consensus bug become the network’s problem. The 33% line matters: below it, Firedancer is a recovery path; above it, every rollout mistake starts looking like an Ethereum-style client diversity fire drill. For apps like Jupiter, Phoenix and Kamino, the alpha is not 1M TPS demos, it’s fewer weird leader/packet-path failures when Solana is already doing ~1k non-vote TPS at low utilization.

Top comment by @Benthic