Drop legacy math parquets superseded by clean test_<benchmark>.parquet + dapo-math-17k.parquet

Reorganize: drop -boxed/-buggy/-fixed variants, add metadata column for slime030 dispatch

Add per-benchmark math eval split: validation_olympiadbench-boxed.parquet

Add per-benchmark math eval split: validation_math500-boxed.parquet

Add per-benchmark math eval split: validation_amc23-boxed.parquet

Add per-benchmark math eval split: validation_minerva-boxed.parquet

Fix upstream control-byte corruption in validation_aime.parquet (boxed-format)

Fix upstream control-byte corruption in validation.parquet (boxed-format)

Fix upstream control-byte corruption in test_other.parquet (boxed-format)

Fix upstream control-byte corruption in test_aime.parquet (boxed-format)

Fix upstream control-byte corruption in test.parquet (boxed-format)

Fix upstream control-byte corruption in dapo-math-17k train (rebuilt from BytedTsinghua-SIA/DAPO-Math-17k, boxed-format)

Upload math_data/validation_aime-boxed.parquet with huggingface_hub

Upload math_data/validation_aime.parquet with huggingface_hub

Upload math_data/validation-boxed.parquet with huggingface_hub

Upload math_data/test_other-boxed.parquet with huggingface_hub

Upload math_data/test_aime-boxed.parquet with huggingface_hub

Upload math_data/test-boxed.parquet with huggingface_hub

Upload math_data/dapo-math-17k-boxed.parquet with huggingface_hub

Add combined test.parquet (test_aime + test_other, 1546 rows)

Add deduplicated math + code data with system instructions