performance

Part 1 of my benchmark best practices series: how I turn a scaling claim into an implied serial fraction with Amdahl’s law, and what the Universal Scalability Law adds. Includes a measured pipeline where assigning more vCPUs per image cut global throughput.