Evidence, not a winner badge
A reproducible web-to-Markdown benchmark
This is a first-party product-team test, not an independent study. We publish the source URLs, retrieval time, generator, raw structural counts, JSON report, and Web2MD Markdown so you can inspect the output instead of trusting a score.
Method and limits
- 1.Public source HTML is fetched once and converted locally with the open-source @web2md/shared engine.
- 2.Jina Reader is requested separately at the same time, so network and upstream rendering can differ.
- 3.Structural counts are descriptive, not quality scores. Download and inspect the Markdown before choosing a tool.
- 4.Authenticated-page access is excluded because it cannot be reproduced without a reader-owned account and browser session.
Run date: 2026-11-05T01:28:38.180Z · Owner: Web2MD product team
Public-page extraction samples
Counts describe output shape; more characters or links do not automatically mean better extraction. Open the files and compare reading order, content coverage, and noise.
| Source | Web2MD output | Structure retained | Jina reference |
|---|---|---|---|
| MDN: HTTP GET HTTP 200 · CC BY-SA 2.5 | Download .md 2,905 chars · 974 ms local conversion | 1 headings · 5 table groups · 3 code blocks · 14 links | HTTP 200 62,962 chars · 9 headings · 534 links |
| Wikipedia: Markdown HTTP 200 · CC BY-SA 4.0 / GFDL | Download .md 36,765 chars · 1196 ms local conversion | 18 headings · 2 table groups · 0 code blocks · 228 links | HTTP 200 47,160 chars · 2 headings · 328 links |
What this benchmark proves
- The exact Web2MD engine can turn two live public sources into downloadable Markdown.
- The output retains document structures that matter for AI input.
- The run is reproducible and failures are visible in JSON.
What it does not prove
- It is not an independent endorsement.
- Two URLs do not predict every website.
- It does not test authenticated access, browser state, or full-site crawling.
- Structural counts are not a quality ranking.