Benchmarks

Benchmarks

RayoMD benchmark claims are dated, scoped, and reproducible. Hardware, storage, fonts, compiler, tool versions, and document shape all affect results.

Current release

• RayoMD 2.6.0 Performance Snapshot - the current Windows fresh-process 2.5 to 2.6 comparison, including hashes, correctness checks, and the strict acceptance caveat.

The README intentionally carries only this compact release-to-release result. Cross-tool comparisons remain dated historical evidence because the tools have different rendering scope and their runtimes age independently.

Specialized reports

• RayoMD tester.md Warm Benchmark 2026-07-17 - warm in-process native generation; excludes startup, input reading, and timed output writing.
• Reversible PDF Feature Benchmark 2026-07-13 - embedding, recovery, hostile-input limits, workflows, output size, and memory.
• RayoMD 2.2.0 Optimization Audit 2026-07-13 - profiling, concurrency, compiler experiments, and release evidence.
• Performance Watcher - maintained commands and regression gates.

Cross-tool archive

• Markdown-to-PDF Speed Comparison 2026-07-13 - RayoMD 2.2.0, Pandoc + XeLaTeX, and md-to-pdf on three synthetic Windows cases.
• Markdown-to-PDF Speed Comparison 2026-07-11 - earlier version of the same three-tool campaign.
• RayoMD vs Pandoc 2026-06-18 - focused local Windows export comparison.
• Commercial Benchmark Campaign 2026-06-06 - large synthetic inputs, batching, and serve mode.

Engineering archive

• Optimization Progress - accepted and rejected native-path experiments.
• Optimization Research Notes - research sources and follow-up candidates.
• Issue 20 Completion Matrix - reversible-PDF completion evidence.

Reading the numbers

• Warm --bench timing measures repeated in-process PDF byte generation.
• Export and batch timing includes process and file I/O.
• Synthetic inputs are not a representative customer corpus.
• Linux/WSL storage matters: native ext4 can be dramatically faster than /mnt/* for many small files.
• Native RayoMD implements a narrower feature set than Pandoc and browser-based renderers. Speed comparisons are not claims of feature or visual equivalence.

Use Performance Watcher for current before/after measurements. Compact version records remain in docs/benchmarks/releases because scripts consume them and they belong to source history.