-
Add Visual_Information annotation column to train/test splits; document in README
-
Add deterministic evaluators (test.raw.json) + correct 5 suspect golds in test.csv
Deterministic string/number/list verifiers replace the LLM judge (229/231, 99.1%). Corrected golds: task3 (+short-beaked echidna), task7 (drop spurious State prefix), task9 (Atomic nucleus), task19 (election), task99 (American Girl dolls). Each env-verified.
-
Upload test.csv
-
Upload test.csv
-
Upload test.csv
-
Upload test.csv
