- Reduce conversion time by 27.9% on the 115-page offline corpus in a controlled
before/after comparison, from 1.732 to 1.250 seconds. Reuse selection statistics,
avoid redundant tree scans and copies, and skip unused code-element rendering. - Preserve Markdown, metadata, and diagnostics across the existing corpus, with
regression coverage for cleanup statistics, JSON-LD precedence, and nested code. - Add reproducible six-library benchmarks and publish timing, memory, import,
footprint, implementation-language, and dependency comparisons in the README. - Keep the package pure Python with zero runtime dependencies. Runtime source grows
by 37 lines (3.3%); benchmark tooling and measurements stay out of distributions.
See the measured comparison for variability,
methodology, and limitations. Performance results do not establish output quality.