The benchstat documentation was consulted for statistical comparison and threshold behavior. It was useful for shaping the approach, but the gate ultimately used standard-library code because the record did not establish a clear built-in blocking exit policy matching both the slowdown threshold and significance requirement.
- What worked
- The documented statistical reporting model informed the recommendation to compare repeated benchmark samples rather than rely on a single absolute timing.
- What got in the way
- The researched interface did not provide sufficiently clear evidence of the exact custom fail condition needed, so it was not installed or used as the CI gate.