- SQL (30 days) says why a value is missing; the generic rank status is shorter
- Rank, status and 30-day reasons that were tooltips are also in the page text
- Model page tile says it shows the point Arena Score of all challenges
- Shorter legend; the rank explanation names the coverage reason
Leaderboards and model pages show the Arena Score with rank ranges
- Leaderboards on the front page and challenge pages rate models by the
Arena Score with its 95 % confidence interval, as a point and a
probabilistic view, with the rank shown as a range
- Models with too few weeks of results or too little coverage show that
status in place of the rank
- MASE (30 days) and SQL (30 days) replace the average columns
- Model pages plot the Arena Score history with its interval and rank range
- About page and track explanation describe the Arena Score
23 days ago
Show each model's track (Reference or Open) on the leaderboards and model pages, with a switch to view one track
Model size was formatted unconditionally, so a model registered without one
crashed the page into the error boundary. Every optional field now falls back
to "Not specified", and a missing publishing date no longer reads as 1970.
Sub-requests are also settled individually, so an unranked model still shows
its details instead of failing the whole page.
Shorten the front page banner and point it at the news section
Participants can register and mint their own API keys in the console, so
/add-model now leads with the self-service path and keeps the email route
as a fallback. The console URL comes from CONSOLE_URL; unset means no
console and every entry point disappears.
Load leaderboard rankings in three requests instead of one per challenge
Below the sm breakpoint the Plotly legend is replaced by an HTML list under
the chart, freeing the full plot width, and drag is disabled so a swipe
scrolls the page instead of drawing a zoom box.
