task_name,version,doi_status,prediction_target,label_source,eligibility_rule,split_strategy,train_rows,validation_rows,test_rows,grade_a_rows,grade_b_rows,metrics,baseline_models,leakage_policy,download_reference_splits,download_task_cards
toxicity_safety_v0_1,20260604_resource_depth_v42,pending_public_archive_before_submission,oligonucleotide safety endpoint triage from molecule/modality/provenance fields,toxicity_endpoint.endpoint_category and endpoint_name,curator_verified accept evidence_grade in A/B,stored_source_plus_molecule_grouped_splits,218,23,22,180,83,AUROC; AUPRC; macro-F1; PCC/Spearman or MSE only for numeric toxicity values,train-majority; modality-prior; evidence-grade-prior; target-prior deterministic baselines,stored splits keep records sharing source identifier and molecule/cohort in one split,/api/download/benchmark_reference_splits.csv,/api/manifest/benchmark_task_cards_v1.csv
offtarget_safety_v0_1,20260604_resource_depth_v42,pending_public_archive_before_submission,observed off-target signal or off-target evidence type,offtarget_evidence.evidence_type and is_computational_prediction,curator_verified accept evidence_grade in A/B,stored_source_plus_molecule_grouped_splits,66,5,10,32,49,AUROC; AUPRC; macro-F1; PCC/Spearman for ranked off-target risk,train-majority; modality-prior; evidence-grade-prior; target-prior deterministic baselines,stored splits keep records sharing source identifier and molecule/cohort in one split,/api/download/benchmark_reference_splits.csv,/api/manifest/benchmark_task_cards_v1.csv
