Measurement data notes

requests.csv: All 4,800 request measurements. Times are relative to the start of each run.
summary.csv: Aggregated values for each condition.
runs.csv: The 12 individual runs and their CPU counter deltas.

duration_s: About 20 seconds measured from the scheduled send start.
cpu_usage_usec_delta: Increase in cpu.stat's usage_usec, in microseconds.
cpu_cores: Recorded CPU usage. CPU observation starts about 0.1 seconds before scheduled sending, so duration_s is not its denominator.
cpu_observation_s_derived: CPU time in seconds divided by the recorded cpu_cores value, reconstructing a denominator of about 20.101 seconds. The original record did not store this duration as a separate field, so this column is not a new independent measurement.
cpu_observation_s_estimate: Difference between two separately recorded timestamps near the counter reads. This is an independent approximation of the CPU observation interval. Recalculating CPU usage with it gives the same three-decimal values shown in the article.
nr_periods_delta / nr_throttled_delta / throttled_usec_delta: Increases in the corresponding cpu.stat counters. Throttled time or period counts are not individual request waiting times or CPU utilization percentages.

Average CPU usage in the article is the mean of the three per-run cpu_cores values. The slower-response measure is the 1,188th of all 1,200 request times sorted in ascending order for each condition, not an average of three run-level values. Existing CSV column names are retained for compatibility.