Should I measure the first output or the completed answer in smart glasses?
Measure both if they represent different useful moments. First useful output tells you when the wearer can begin understanding the result. Completion tells you when the requested answer or action is finished. A spinner or acknowledgment is feedback, but it is not the answer.
Define what counts as useful
A streaming response can start quickly yet delay the essential information. For a translation, the first word may not reveal the meaning. For an agent task, “Working” confirms receipt but says nothing about completion. The metric should match what the wearer can actually do with the output.
Report time to acknowledgment, time to first useful content and time to completion separately when all three matter. That makes responsiveness visible without allowing a quick animation to stand in for the slower underlying task.