Measure both raw tokens generated and productive tokens: tokens that produced merged work, closed issues, or verified receipts. Report cache hit rates, input vs output split, and throughput per provider (community numbers suggest ~26 TPS, ~98% input, ~87% cache hit). Use this to rebalance lanes toward productive work.
Measure both raw tokens generated and productive tokens: tokens that produced merged work, closed issues, or verified receipts. Report cache hit rates, input vs output split, and throughput per provider (community numbers suggest ~26 TPS, ~98% input, ~87% cache hit). Use this to rebalance lanes toward productive work.