{"id":"W4412378237","doi":"10.1145/3726302.3730244","title":"Response Quality Assessment for Retrieval-Augmented Generation via Conditional Conformal Factuality","year":2025,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"Universitas Brawijaya","keywords":"Computer science; Conformal map; Quality (philosophy); Information retrieval; Artificial intelligence; Mathematics; Physics","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":[0.01457145,0.00145087,0.001407784,0.002300023,0.0009100031,0.002312816,0.00212728,0.00203618,0.006667965],"category_scores_gemma":[0.08909014,0.0004564423,0.001182702,0.001132612,0.001437132,0.003729865,0.00384706,0.002046807,0.0024178],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001360645,"about_ca_system_score_gemma":0.001912739,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002562408,"about_ca_topic_score_gemma":0.002736431,"domain_scores_codex":[0.9839873,0.008102507,0.0008531554,0.002589744,0.003919589,0.0005477223],"domain_scores_gemma":[0.9348242,0.04275153,0.002806884,0.01242624,0.006215295,0.0009758649],"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":[0.002984524,0.0006897045,0.0246689,0.001566148,0.0003382923,0.0005931268,0.002573661,0.1174504,0.03503087,0.0308068,0.02707157,0.7562259],"study_design_scores_gemma":[0.0001879933,0.0005584406,0.005234239,0.0001001391,0.0001460694,0.000450357,0.0003530046,0.9278136,0.02044015,0.03486518,0.009743341,0.0001074252],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1204956,0.001448891,0.8429873,0.001304364,0.0002949911,0.0009770941,0.002032293,0.02354455,0.006914996],"genre_scores_gemma":[0.7729212,0.0002294853,0.2181111,0.0005103849,0.0001915574,0.0005161298,0.003785717,0.001014535,0.002719829],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01457145,"threshold_uncertainty_score":0.07706207,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08704555234092608,"score_gpt":0.3904801283648089,"score_spread":0.3034345760238829,"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."}}