Definition
Plain language
How much useful work a system actually finishes within deadlines.
As stated in the literature
In serving systems, throughput restricted to requests that complete successfully within latency targets.
Why it matters: Real users care about completed work within deadlines, not raw request counts, so goodput is the metric that maps to actual product quality.
For example, a server that handles a thousand requests per second but lets a third time out has much lower goodput than its raw throughput suggests.
Heard on the show
“The paper draws a hard line between throughput and goodput.”Episode 016 — Why Your Coding Agent Stalls While the GPU Runs Hot