Benchmark Radar Day 13: Dashboard Polish and Evidence Grounding
Published:
Every claim now links to its source. Day thirteen added evidence grounding, daily Q&A, and several UX fixes.
Author: Koutian Wu; GitHub: ktwu01
What shipped
Favicon. The site finally got an icon. Small detail, big identity signal.
URL parameter scoping. Filters are now written only for the active view into the URL. Previously, all views’ filters leaked into the URL, creating confusion when switching between views.
Daily Q&A on dashboard. The daily Q&A section is now rendered on the dashboard, making the radar’s self-questions and answers visible to users.
Briefing evidence grounding. The daily briefing now answers questions instead of only listing findings. Each answer is grounded in specific evidence sources. 80% of previously discarded briefing evidence was recovered.
KW-Bench rubric owned. The KW-Bench rubric is now maintained within this repository instead of being an external dependency.
Benchmark builder survey. A survey of who builds the popular benchmarks was added (#164).
Trajectory points inspectable. Points on the trajectory chart can now be clicked or focused to reveal details.
Complete contact inventory. The benchmark contact CSV export now includes the complete inventory.
OpenAlex null title fix. Untitled OpenAlex works no longer crash the daily run. They are skipped with a warning.
Why it matters
Evidence grounding was the most important change of the day. A radar that says “benchmark X is trending” without citing where that claim comes from is just opinion. A radar that says “benchmark X appeared in 3 model cards this week (GPT-4 card line 42, Claude card line 18, Gemini card line 31)” is evidence.
The URL parameter scoping fix addressed a real user confusion: when you shared a URL, it carried filters from views the recipient was not looking at. Scoping filters to the active view made URLs actually represent what they showed.
Issues addressed
- #160: URL parameter view scoping and favicon
- #161: OpenAlex null title fix
- #162: briefing evidence starvation
- #159: render daily Q&A
- #163: trajectory interactions
- #164: benchmark builder survey
- #166: render daily questions on dashboard
- #167: complete contact CSV
- #168: trajectory point inspection
Day fourteen: feed coverage, briefing reliability, and production Q&A.