How many trials do I need for a smart-glasses latency test?
There is no single trial count that makes every benchmark reliable. You need repeated measurements under the conditions you want to describe, with the sample size reported. A small repeatable pilot can reveal variation and help decide where a larger test is needed.
Sample the conditions, not just the same easy moment
Warm and cold starts, phone background state and different network conditions can produce different results. Mixing them without labels makes the result difficult to interpret. Repeating only the fastest condition many times does not establish performance in the others.
Begin with a clearly defined workload and preserve every attempt, including failures. If results vary substantially, expand the sample and investigate why. A single best run is useful as a demonstration of possibility, not as a description of typical performance.