Topic
How performance measurements are designed, ordered, aggregated and validated, including randomised or round-robin sweeps and median aggregation.
A week of measurements on one Mac mini puts the lru_cache wrapper around strptime at 3x when dates repeat and slightly slower when they do not. A Polars 2.0 release candidate finishes the same report in 0.13 s.
Reality
A JMH sweep at 1, 2, 4, 6 and 8 threads on four cores puts a synchronized counter ahead of a hand-rolled compare-and-swap loop once every thread shares one instance, and the confidence intervals never overlap.
Reality
Django-Bolt's published figure comes from an endpoint that returns a fixed response, and a devops.com analysis argues the number a team actually needs is one from its own endpoint with authentication and database queries still in the path.
Reality
The rule for choosing a hosted PDF API or a local library for shipping labels rests on p95 and p99 measured under identical load, and the percentile helper the post prints reports whichever rank the sample count lands on.
Reality
The daemon's prompt_eval_count includes cache reads and prompt_eval_duration does not, so one qwen2.5:7b prompt reported 13,826 tokens a second and 43 tokens a second half a minute apart. One subtraction fixes it.
Reality
A developer discarded a full data set after a sanity re-run disagreed by up to 54%. The problem was measurement order on a passively cooled chip, not noise.
Reality