Benchmark Radar Day 42: The Recent-Releases Ranking Reaches the Dashboard

Day forty-two of Benchmark Radar. The ranking engine for recent releases moves from backend to the dashboard, and the daily radar snapshot captures the first full week of ranked cohorts. Scoreboard: 165 stars, 29 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.

A cohort is a set of releases grouped by when they were published. A window is the span of days a release must land in to count. The ranking engine evaluates seven-, thirty-, and ninety-day windows with the data contract validated at ingest.

PR #531 phase two wires the ranking engine into the dashboard. The Today page now shows a ranked feed of recent releases grouped by cohort window, with each entry displaying its benchmark identity, normalized score, test protocol, and evidence link. Releases without enough durable signal show a limited_signals badge instead of a rank number. The seven-day cohort updates daily, the thirty-day cohort weekly, and the ninety-day cohort monthly.

PR #562 adds a protocol column to the ranked feed so readers can see exactly which measurement rules produced each score. PR #563 fixes a hydration mismatch where the cohort label rendered differently on server and client. PR #556 (from Day 41) brought the sunblaze-ucb organization into the discovery registry, and today’s scan picked up their first benchmark.

Why this matters.

A ranking that lives only in a database is a promise. Putting it on the dashboard makes the rules visible: readers can see which cohort a release belongs to, which protocol measured its score, and whether it earned a rank or a limited_signals label.

The cohort structure turns a flat list into a timeline. A benchmark that appeared in the seven-day window but not the thirty-day window tells a different story than one that persists across all three.

Issues addressed

• #530: data contract and ranking engine for recent releases, phase two (dashboard integration)
• #562: protocol column in ranked feed
• #563: hydration fix for cohort labels
• scoreboard: 165 of 1,000 stars, 29 forks

Day forty-three: paper preparation intensifies — source counts, figure data, and the audit pipeline freeze for the arXiv submission.

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