v2.9.4

Changed

• Keep the benchmark score labels naming both splits, and decode test output safely
• Cut jargon and parentheses from the four tabs
• Rewrite the Dataset, Design, Benchmark and Element data instructions simply
• Correct three factual errors in the dataset cards and abstracts
• Document the public-text rule for agents

Full Changelog: v2.9.3...v2.9.4