docs: explain benchmark summary statistics

README.md

@@ -44,6 +44,18 @@ If you want to compare the runtimes of different programs, you can pass multiple
 hyperfine 'hexdump file' 'xxd file'
 ```
 
+### Interpreting benchmark results
+
+The summary for each command combines measurements from all timed benchmark runs.
+`Time (mean ± σ)` reports the mean elapsed (wall-clock) time and the corrected
+sample [standard deviation](https://en.wikipedia.org/wiki/Standard_deviation) `σ`
+of those times. A smaller `σ` means that the measured runtimes were more tightly
+clustered around the mean.
+
+`User` and `System` report the mean CPU time per run spent in user mode and
+kernel/system mode, respectively. They are CPU-time components and do not need
+to add up to the elapsed time.
+
 ### Warmup runs and preparation commands
 
 For programs that perform a lot of disk I/O, the benchmarking results can be heavily influenced

doc/hyperfine.1

@@ -81,6 +81,18 @@ A command\-line benchmarking tool which includes:
 .RS
 * Parameterized benchmarks (e.g. vary the number of threads)
 .RE
+.SH BENCHMARK OUTPUT
+The summary for each command combines measurements from all timed benchmark
+runs.
+.PP
+\fBTime (mean \(+- sigma)\fP reports the mean elapsed (wall\-clock) time and
+the corrected sample standard deviation, sigma, of those times. A smaller
+standard deviation means that the measured runtimes were more tightly
+clustered around the mean.
+.PP
+\fBUser\fP and \fBSystem\fP report the mean CPU time per run spent in user
+mode and kernel/system mode, respectively. They are CPU\-time components and
+do not need to add up to the elapsed time.
 .SH OPTIONS
 .HP
 \fB\-w\fR, \fB\-\-warmup\fR \fINUM\fP