{"id":"W4413953529","doi":"10.14778/3742728.3742753","title":"Robust Plan Evaluation Based on Approximate Probabilistic Machine Learning","year":2025,"lang":"en","type":"article","venue":"Proceedings of the VLDB Endowment","topic":"Rough Sets and Fuzzy Logic","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"IBM (Canada); University of Ottawa","funders":"","keywords":"Plan (archaeology); Probabilistic logic; Computer science; Machine learning; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006307388,0.001183584,0.002137791,0.001603832,0.0005285119,0.002485813,0.002313519,0.001244019,0.001850634],"category_scores_gemma":[0.02298528,0.0007103962,0.001155116,0.001666223,0.001688645,0.004080151,0.002439599,0.002013884,0.0003614438],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003001285,"about_ca_system_score_gemma":0.002764692,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006202673,"about_ca_topic_score_gemma":0.005366314,"domain_scores_codex":[0.9929364,0.002476918,0.0004358171,0.001108145,0.002600735,0.0004418762],"domain_scores_gemma":[0.9865859,0.009171231,0.001262878,0.001532779,0.001159896,0.0002873334],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000307214,0.00009438382,0.001888332,0.0001289358,0.00009609753,0.00005837588,0.0001192504,0.865527,0.002330475,0.0223054,0.001577393,0.1055671],"study_design_scores_gemma":[0.000009477503,0.00003533204,0.0001161908,0.000006231203,0.000008492429,0.00001137075,0.00001048179,0.9904697,0.0007410938,0.00838417,0.0002009188,0.000006488992],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02679198,0.00063153,0.9696519,0.0003764669,0.00001919898,0.0001166753,0.000159651,0.001276024,0.0009765781],"genre_scores_gemma":[0.672756,0.0003355579,0.3244988,0.0002280316,0.00007129122,0.0002346067,0.0005174961,0.0002511801,0.001106979],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.006307388,"threshold_uncertainty_score":0.03335708,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03655889814260427,"score_gpt":0.2401569082094122,"score_spread":0.2035980100668079,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}