Docs: Average column is item-weighted mean

src/about.py

@@ -45,7 +45,7 @@ The **Arabic LLM Leaderboard** measures how well large language models understan
 a **synthetic MSA** track and a **webscraped + aptitude** track — graded by strong LLM judges.
 
 Each submission is scored on **13 capability and domain dimensions** (7 synthetic capability categories and
-6 webscraped domains/groups); the **Average ⬆️** column is the mean across all 13. See the **📝 About** tab for
+6 webscraped domains/groups); the **Average ⬆️** column is the **item-weighted** mean across all 13 (each dimension weighted by its number of benchmark items, so every item counts equally). See the **📝 About** tab for
 the full dataset composition, scoring methodology, and how to reproduce results, and the **🚀 Submit** tab to add
 your own model.
 """
@@ -70,7 +70,7 @@ percentages.
   the metric is a single judged-accuracy score per dimension.
 - The two tracks use different solver/judge pairs (see table above) and their scores are **kept per track**.
 - The leaderboard's per-model table breaks each model down into **13 dimensions** plus an **Average ⬆️** (the
-  mean of all 13). A model must report **all 13** dimensions to appear on the board.
+  item-weighted mean of all 13 -- each dimension weighted by its item count). A model must report **all 13** dimensions to appear on the board.
 
 ## 1) By format (combined item counts; scores kept per track)