compare_runs
ChatGPTCompare two CodSpeed performance runs and return a markdown performance report. Comparing a local or Wizard run compares only the benchmarks each run measured: a benchmark only one run measured is listed as new or missing. Between two CI runs, a benchmark a run skipped is compared using the result it carried over from an earlier run, as on the CodSpeed compare page. Shows benchmark-level comparisons including improvements, regressions, new/missing benchmarks, and formatted values. When hardware or software environment differences are detected between the two runs (e.g., different CPU, OS, language runtime version, or linked libraries), an "Environment Differences" section appears first in the report. Each difference lists exactly the benchmarks whose base and head ran across it and explains inline how that kind of difference affects each instrument; benchmarks compared on matching runners are explicitly called out as unaffected. Rely on this attribution rather than general intuition about which instrument an environment change 'should' affect: it reflects where the runners actually differed. When the two runs were measured in ways that make their results not directly comparable (e.g., a runner major-version bump that changes how measurements are computed), only a "Runs Not Directly Comparable" warning is returned — the summary and per-benchmark results are omitted because the comparison would be misleading. The report header includes
comparison_url, the CodSpeed page for this comparison. Give it to the user when they want to look at the results themselves, for example to open flame graphs or to share the comparison. Use this to understand the performance impact between two runs (e.g., a PR head vs its base).
