Benchmark · Measured
Main-content extraction quality
On 511 held-out WCXB pages, Webfold is level with Trafilatura. Webfold scores F1 0.861. Trafilatura 2.3.1 scores 0.863. Its edge is speed. It is not accuracy.
Webfold on the public WCXB benchmark. 511 held-out pages. Scored with the benchmark's own tool.
What question does this benchmark ask?
How much main content does the extractor recover? How much junk does it leak?
How was it run?
- WCXB: 2,008 human-reviewed pages from 1,613 domains in seven page types; 511 held out as a test set.
- Metric: word-level F1 against each page's plain-text main content, scored by the benchmark's unmodified evaluation script.
- Output is converted from Markdown to plain text before scoring.
What are the results?
| System | Test F1 |
|---|---|
| rs-trafilatura (published) | 0.893 |
| Trafilatura 2.3.1 (re-run here on the same 511 pages) | 0.863 |
| Webfold, with trained main-content selection | 0.861 |
| Trafilatura (published in the benchmark README) | 0.833 |
| Webfold, before selection | 0.815 |
| dom-smoothie (published) | 0.808 |
| Readability (published) | 0.726 |
On accuracy Webfold is level with Trafilatura, not ahead of it: Trafilatura 2.3.1 scores 0.863 when we re-run it on the same pages, and the 0.833 in the benchmark's README does not reproduce. The best published system, rs-trafilatura, is ahead at 0.893. Webfold's case is speed and memory (the HTML conversion page), with accuracy in the same band. Its strongest page types are documentation (0.927) and articles (0.920).
Recall is high (0.931) and precision is the weaker side (0.834).
What does this not show?
Know the limits
- Trafilatura was re-run here (version 2.3.1: 0.863; pinned to 2.0.0 it gives 0.826); the other systems' scores are the benchmark's published baselines, not re-run.
- The trained selector was fitted on the development split only; the test split was scored once.
- Firecrawl's extractor has not been scored on this benchmark.
Which other benchmarks are there?
- HTML to Markdown conversion Measured
- Document conversion Measured
- Web archive extraction Measured
- Footprint of a scrape or crawl Directional
- PII detection against Presidio Measured
- Localisation extraction Directional
- PDF text extraction with OCR Directional
- PDF extraction against other extractors Directional
- LaTeX conversion with BibTeX Directional
- A whole site to Markdown Directional
- Agent-browser compatibility (Lexbench) Measured
- Headless engine over CDP Directional
- Search-results parsing Directional
- Folder search Directional
Check it on your own pages.
Run the command-line tool on your content.