Benchmarked function is run one more time than specified

Consider the following example:

first_run = false
function f()
  global first_run
  if !first_run; first_run = true; else sleep(1); end
  println(time())
end
@benchmark f() evals=1 samples=1 

which outputs
1.734518515199018e9
1.734518516201369e9
BenchmarkTools.Trial: 1 sample with 1 evaluation.
Single result which took 1.002 s (0.00% GC) to evaluate,
with a memory estimate of 624 bytes, over 11 allocations.

It seems that `f` is run an additional time before the actual sample is taken. This seems not to be documented. I wonder if this is to even compilation time out. If yes, where do I find information about that? In particular, is it possible to skip this, or trigger compilation with a cheaper instance?

Run on Julia 1.10 with BenchmarkTools v1.5.0.

Start with the @benchmark entry point and the evals=1, samples=1 configuration shown in the report. Reproduce the extra invocation on Julia 1.10 with BenchmarkTools v1.5.0, then trace how setup, compilation, and sampling are handled. Done means explaining the additional run and documenting or exposing the relevant behavior, including whether it can be skipped.