How do I make a smart-glasses benchmark reproducible?
State the device, firmware, app version, network conditions, workload and exact start and finish events. Preserve representative raw results with private information removed. Another person needs enough detail to repeat the procedure and understand which conditions the findings cover.
Describe the workload as carefully as the hardware
A short cached response and a new long request may exercise different paths on the same device. Phone background state, connection setup and whether models or assets are already loaded can also matter. Keep those conditions explicit so later comparisons measure a real change.
Publish the procedure alongside the result and distinguish observed outcomes from untested scenarios. A reproducible desk test is valuable on its own; it becomes less useful when its conclusion expands to all-day use or hardware combinations that were never measured.