4 release tags and a pinned dev-main when its measured source differs from the newest tag. Candidate PR points are included when requested. Lower elapsed time is faster. These measurements use a shared host; inspect sample spread before treating small changes as regressions.
Raw samples and provenance · Download CSV · Measurement report
Relative mode gives each case its own oldest-tag baseline of 1×. Equal starting ratios do not mean equal milliseconds; select Milliseconds to see elapsed time. Graphs use a logarithmic vertical scale. Filled points match oldest-tag output. Hollow points have different output; lines connect only matching adjacent outputs. The html_table and html_definition_list cases measure HTML import. Tag spacing shows version order, not elapsed calendar time. Each engine has its own release tags and runtime. PHP runs with CLI opcache/JIT off; Rust uses optimized release builds. These timings exclude process startup and belong to a separate measurement from the headline benchmark.