Skip to content

Releases

v0.1.0: multilingual and AI-search SEO auditor, with a trap site and its answer key

Choose a tag to compare

@thibaultbordrie-cloud thibaultbordrie-cloud released this 09 Nov 15:25
139e38b

First public release. One Rust binary, no runtime to install.

What it does

  • 38 rules for sites in French, Japanese and Chinese: hreflang (invalid codes with the intended fix, return links, self links, targets in error, language or Chinese-script mismatch), html lang contradicted by the text, zh-Hant vs zh-Hans, title width with CJK characters counted double, canonicals pointing to another language, broken links, redirect chains, sitemap errors.
  • AI search (GEO): crawlers that AI assistants use to cite pages (OAI-SearchBot, Claude-SearchBot, PerplexityBot) blocked in robots.txt, told apart from training crawlers (GPTBot, CCBot, Google-Extended: notice only); JavaScript-only pages; llms.txt; JSON-LD validity and Product offers.
  • A trap site with its answer key: kensa fixture writes Lune & Lin, a fictional shop in three languages (41 pages) with 25 planted defects; the seed moves them. kensa score gives precision, recall and F1 for any tool.

Results

F1 = 1.00 on 100 seeds (every defect found, no false alarm), in about 2 seconds. CI on Linux, macOS and Windows.

Get it

Binaries for Linux (x86_64), macOS (Apple silicon) and Windows (x86_64) are attached below by the release workflow. Or:

cargo install --git https://github.com/thibaultbordrie-cloud/kensa
kensa bench

All names, products and texts in the trap site are fictional. MIT licence.