A benchmark needs a stable workload and a result that remains observable. In a normal Go package a Benchmark function receives a testing.B, repeats the operation and runs with go test -bench=. Report ns/op and allocation data with the workload and environment, then repeat enough times to understand noise. A CPU profile answers where execution time was sampled; a heap profile answers a different storage question. The editor can run testing.Benchmark and testing.AllocsPerRun directly. It does not run the profiling tools or the race detector. Two tasks use fictional measurement records; two run actual testing APIs with deterministic assertions. Calculate per-operation values, compare equal workloads and avoid treating cumulative time as exclusive work. A smaller number from a different workload is not a valid improvement claim.
Convert before division.