Benchmark Radar Day 35: One URL for Every Benchmark

Day thirty-five of Benchmark Radar. A week-old copycat outranked the real project for its own name because search engines could see only four pages here. Today every one of the 1,173 benchmarks the radar tracks has a page of its own. Scoreboard: 123 stars, 23 forks.

Author: Koutian Wu; GitHub: ktwu01

Benchmark Radar is live. Track new AI benchmarks, datasets, and leaderboards every day: open the dashboard or star it on GitHub.

SEO means making a website easy for search engines to find and rank. Indexable means a page a search engine can read and list. Structured data is machine-readable description embedded in a page, and a sitemap is the list of URLs a site wants search engines to know about. A ledger is a public list that records who earned what, and a byline is the list of authors on a document.

PR #442 is the big one. Each benchmark now gets a static page with a unique title, description, canonical URL, structured data, and per-source score tables, plus a directory page. The sitemap grows from 4 to 1,177 URLs. Internal links become absolute so analyzers can find them, the homepage gains Organization and enriched WebSite and Dataset schema, llms.txt is published, and the published CSS is minified. The rendered dashboard does not change.

PR #443 fixes the one thing that broke: the interactive view link on those pages now points at the leaderboard, where a frontier-model slug can actually render.

PR #449 publishes the contributor score ledger. Contributors can claim an issue, a claim expires after seven days, one person holds an issue at a time, and a daily job keeps the ledger current. PR #450 makes sure only work a contributor can actually complete is scored, and PR #451 raises the collaborator threshold from 6 to 12 points, with every accepted real use case worth 6. PR #458 links six scored paper research issues, each worth 6 to 12 points.

PR #441 adds a citation for the technical report and simplifies the local setup instructions, and PR #444 thanks Xiaopai Liu for the shout-out, moves the acknowledgements to the end, and flags the CLI version.

Why this matters.

The data advantage was real but invisible. A project holding 1,173 evidence-linked benchmarks was exposing four indexable URLs, so a copycat with none of the data could rank higher for the brand name. One page per benchmark turns the dataset itself into the search surface, which is exactly the surface a copycat cannot copy.

Outside contributors need rules they can trust. A public ledger, a claim mechanism, and scores tied to work that is actually completable turn contribution from a favor into a job with a fair price.

Issues addressed

• SEO: one static page per benchmark, absolute links, and structured data
• #440: README acknowledges Xiaopai Liu and moves acknowledgements last
• #449: contributor score ledger with claiming and seven-day expiry
• #450: only contributor-completable work is scored
• #451: collaborator threshold raised to 12 points
• #458: six scored paper research issues linked
• scoreboard: 123 of 1,000 stars, 23 forks

Day thirty-six: a frontier model card with 37 scores, and a search that shows its work.

Want to follow Benchmark Radar? Star the repo on GitHub for daily updates, or open the live dashboard to explore the scans.