Skip to content

ALL-PERF1003 · benchmark_quality

Judge whether a benchmark supports the decision made from it.

This is a contextual rule for all languages. Read its implementation.

Compare the decision, workload distribution, data size, environment, warmup, repetitions, variance, baseline, resources, end-to-end effects, and recency.

Findings cite benchmark code, commands, environments, samples, statistics, and the claimed decision.

Focused microbenchmarks are valid for isolated mechanism questions with bounded claims.

A representative service workload with controlled comparisons is representative. Timing one parser token and claiming end-to-end speedup is micro_only.

  • Cites “Systems Performance”
  • Cites “The Python Standard Library”, timeit
  • Cites “Beyond the Basic Stuff with Python”, Measuring Performance and Big O