On one trial you may be fully focused and perfectly prepared. On the next, your attention may drift for a fraction of a second. Tiny differences in motor preparation and input/display timing can also change the result.
This variability is expected, not proof that the timer is randomly broken.
If you take enough trials, eventually you may produce an unusually fast response. Reporting only that minimum makes performance look better than what you usually reproduce.
For self-tracking, use your median or average as the primary number and keep the best as a fun secondary stat.
The median is the middle value after sorting results. One very slow distraction has less influence on the median than on the mean, which makes the median useful when a small set contains an obvious slow tail.
The mean remains useful too; the important thing is knowing what each summary emphasizes.
Two players can have the same 220 ms average while producing very different distributions. One might cluster tightly around 215–225 ms. Another might alternate between 180 ms and 260 ms.
Consistency can therefore reveal a change that the average alone hides.
Attention, anticipation, movement timing and hardware scheduling all vary. Larger differences can also happen when a trial contains a lapse or distraction.
Not automatically. Slow trials are part of performance unless there was a clear external interruption or invalid trial.
Neither is universally better. Median is less affected by extreme slow values; average uses every value and is common in research. Looking at both is useful.