{"id":"W4407356147","doi":"10.1145/3709719","title":"Reliable Text-to-SQL with Adaptive Abstention","year":2025,"lang":"en","type":"article","venue":"Proceedings of the ACM on Management of Data","topic":"Distributed and Parallel Computing Systems","field":"Computer Science","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University; University of Toronto","funders":"","keywords":"Computer science; SQL; Programming language; Database","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":[0.007771262,0.001265395,0.0009940852,0.0007792821,0.0005581075,0.002339134,0.004209791,0.001734918,0.003401929],"category_scores_gemma":[0.04146148,0.0007673644,0.0009996502,0.0006618293,0.001901956,0.00630832,0.00526101,0.002695811,0.001779571],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009060062,"about_ca_system_score_gemma":0.001545325,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002212956,"about_ca_topic_score_gemma":0.002502219,"domain_scores_codex":[0.9917423,0.003296509,0.0005446021,0.001386677,0.002748844,0.0002812516],"domain_scores_gemma":[0.9718204,0.01443824,0.001441899,0.009617512,0.002227117,0.0004548638],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00268958,0.0008219838,0.01060277,0.0005317362,0.0001887054,0.0007908033,0.001968712,0.3637539,0.06737945,0.05882824,0.0196393,0.4728049],"study_design_scores_gemma":[0.00005226412,0.000112863,0.0002360229,0.00000934358,0.00001492883,0.00008006486,0.0000422573,0.9584796,0.01169343,0.02653138,0.002721254,0.00002650214],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03966335,0.0001518277,0.9179853,0.0004808357,0.00004185028,0.0001809628,0.0002729658,0.03987299,0.001349997],"genre_scores_gemma":[0.6025104,0.0001184928,0.3892972,0.0005299482,0.00007492406,0.00030052,0.000950656,0.00318349,0.003034322],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.007771262,"threshold_uncertainty_score":0.04109889,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03721397291859412,"score_gpt":0.2741668219245275,"score_spread":0.2369528490059334,"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."}}