{"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":"codex-gemma-dda1882f352a","candidate_categories":["open_science"],"consensus_categories":[],"category_scores_codex":[0.0004960598,0.0001083104,0.0001585676,0.0001072813,0.00007203928,0.0000819789,0.008341542,0.00002164034,0.000001294886],"category_scores_gemma":[0.00004974271,0.00007304428,0.00003100393,0.0008598426,0.00002872769,0.0003657976,0.006209785,0.00006846543,0.00001036481],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000207606,"about_ca_system_score_gemma":0.00001339284,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001369832,"about_ca_topic_score_gemma":4.731428e-7,"domain_scores_codex":[0.9988154,0.000004866989,0.0002378821,0.0004100239,0.0003725998,0.0001592177],"domain_scores_gemma":[0.9979588,0.00002589425,0.0002235868,0.001624074,0.0001429893,0.00002468381],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001851634,0.0003920651,0.003737384,0.001412963,0.0005635452,0.000001731319,0.0002895336,0.001469888,0.0006248912,0.7292243,0.2361812,0.02591744],"study_design_scores_gemma":[0.009586477,0.004309912,0.1990767,0.03966447,0.001219126,0.00001859644,0.006592786,0.1678176,0.01924423,0.1537234,0.3959917,0.002755016],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.1333856,0.0002984839,0.390639,0.02994033,0.001873072,0.004966131,0.0003803947,0.000633179,0.4378838],"genre_scores_gemma":[0.927381,0.000008973928,0.06936678,0.0001661547,0.00001443073,0.00001055705,0.000009666273,0.000005616913,0.00303685],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7939954,"threshold_uncertainty_score":0.9970238,"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."}}