View
Normalize shows time per element. Spread overlays per-point error
bars (min · median · max); click any point to inspect its individual
samples.
How a curve is computed
Each independently generated random array is one sample. For
every sample, the minimum time over repeated timing
measurements is kept — stripping OS jitter, GC pauses and context
switches. At each axis value, each point is the median across
those independent samples. Only the sort itself is timed; I/O,
parsing and allocation are excluded.
Switch Per-array reduction to Warm-up to keep only the
last timing repetition for each sample instead of the minimum —
emulating warm-up benchmarking, where the final measurement is the
warmed-up result.