Each graph displays benchmark results relative to its baseline commit, which is the latest stable release (e.g., 1.18.3) at the time of testing. The 95% confidence interval is displayed in light gray. On hover, the graph displays the benchmarked commit at that point (click to view full commit).
Note that some commits are not tested, so there could be
multiple commits (not shown) between two points on the graph.
See the gotip-linux-amd64_debian12-perf_vs_release
column on the
build dashboard.
Individually tested commits have their own green box. A single
box that covers multiple commits indicates that only the latest
commit in the range was tested.
The 'Go branch' selection above is the Go branch that benchmarking ran against on the build dashboard.
Using the 'unit' search box above leads to a page that shows much more detail about each individual point in the selected time range. Another way of reaching this page is to click the unit name that acts as each graph's title, i.e. "sec/op".
Further documentation is available on the wiki.