{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":5,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":5,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"query_hash":"d5cdbf56fc71","filters":{"venue":"The Art Science and Engineering of Programming"}},"results":[{"id":"W2899691573","doi":"10.22152/programming-journal.org/2019/3/4","title":"Entombed: An archaeological examination of an Atari 2600 game","year":2018,"lang":"en","type":"article","venue":"The Art Science and Engineering of Programming","topic":"Digital Games and Media","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"Natural Sciences and Engineering Research Council of Canada; Strong","keywords":"Computer science; Code (set theory); Reuse; Table (database); Video game; Variety (cybernetics); Excavation; Artificial intelligence; Programming language; Archaeology; Engineering; Multimedia; History; Database","retraction":null,"screen_n_in":null,"score":{"opus":0.02313949228205194,"gpt":0.2926490655064982,"spread":0.2695095732244462,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00201567,0.0000457278,0.00007299164,0.00006389817,0.00006919222,0.00004575627,0.0002829481,0.0000217882,0.00000577996],"category_scores_gemma":[0.0003311316,0.00003225695,0.00001233832,0.0003831951,0.001898514,0.0005236687,0.00006352034,0.00003970072,0.00000149537],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001329839,"about_ca_system_score_gemma":0.0000507747,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007129961,"about_ca_topic_score_gemma":0.00009593999,"domain_scores_codex":[0.999166,0.00001872528,0.00009889355,0.0001222124,0.0003709613,0.0002232391],"domain_scores_gemma":[0.9995699,0.0000380573,0.00003808446,0.0001186567,0.0001278206,0.0001075146],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000004088778,0.00005204113,0.0004760058,0.00001158171,0.000002398704,6.481378e-7,0.0570408,0.00003878015,0.008498612,0.0214762,0.000001553153,0.9123973],"study_design_scores_gemma":[0.0008470522,0.005113092,0.2952448,0.0002916116,0.00006363009,0.00002167154,0.05320349,0.0547323,0.04134414,0.00292318,0.5452886,0.0009265008],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9967129,0.00002025588,0.0005401822,0.00008432778,0.00008876985,0.00009779351,3.946176e-7,0.00003274644,0.002422662],"genre_scores_gemma":[0.9964685,0.000008839792,0.003288744,0.00001068762,0.0000836309,0.00000360014,6.688686e-7,0.000002453781,0.0001328717],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9114708,"threshold_uncertainty_score":0.6995152,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4407066782","doi":"10.22152/programming-journal.org/2025/10/13","title":"PolyDebug: A Framework for Polyglot Debugging","year":2025,"lang":"en","type":"article","venue":"The Art Science and Engineering of Programming","topic":"Model-Driven Software Engineering Techniques","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"Agence Nationale de la Recherche","keywords":"Polyglot; Debugging; Computer science; Programming language","retraction":null,"screen_n_in":null,"score":{"opus":0.009699444165104274,"gpt":0.2601179501833422,"spread":0.2504185060182379,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001024675,0.0001204957,0.0001409862,0.0002450606,0.000217798,0.00015793,0.001028975,0.00003825606,2.341128e-7],"category_scores_gemma":[0.0002079845,0.00009485894,0.00004067035,0.001079548,0.0001643134,0.0003557689,0.0003422834,0.0001220225,5.345033e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003841447,"about_ca_system_score_gemma":0.00009617866,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00000762977,"about_ca_topic_score_gemma":3.418412e-7,"domain_scores_codex":[0.9989209,0.000005183595,0.0001778144,0.0002751618,0.0002425905,0.0003783059],"domain_scores_gemma":[0.9990965,0.0002153581,0.00004162619,0.0004739456,0.0001146579,0.00005793146],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000001542448,0.000009847864,0.0001177622,0.00005751766,0.000008751197,3.701249e-7,0.0003436347,0.000983774,0.003500198,0.781556,0.00005860515,0.213362],"study_design_scores_gemma":[0.00019059,0.0001184702,0.00217775,0.0005558296,0.00002333208,0.0000145618,0.00002048694,0.8669259,0.03282997,0.01914206,0.07763707,0.0003639666],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.004578942,0.0002759286,0.9930724,0.0009965473,0.0001907192,0.000275938,4.114478e-7,0.000575794,0.00003332926],"genre_scores_gemma":[0.391309,0.000005743696,0.6085386,0.00005454194,0.00001374371,0.00005395761,9.705798e-8,0.000005007151,0.00001934659],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8659421,"threshold_uncertainty_score":0.3868236,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4408014634","doi":"10.22152/programming-journal.org/2026/10/13","title":"PolyDebug: A Framework for Polyglot Debugging","year":2025,"lang":"en","type":"article","venue":"The Art Science and Engineering of Programming","topic":"Model-Driven Software Engineering Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"","keywords":"Polyglot; Debugging; Computer science; Programming language","retraction":null,"screen_n_in":null,"score":{"opus":0.009699444165104274,"gpt":0.2601179501833422,"spread":0.2504185060182379,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001024675,0.0001204957,0.0001409862,0.0002450606,0.000217798,0.00015793,0.001028975,0.00003825606,2.341128e-7],"category_scores_gemma":[0.0002079845,0.00009485894,0.00004067035,0.001079548,0.0001643134,0.0003557689,0.0003422834,0.0001220225,5.345033e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003841447,"about_ca_system_score_gemma":0.00009617866,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00000762977,"about_ca_topic_score_gemma":3.418412e-7,"domain_scores_codex":[0.9989209,0.000005183595,0.0001778144,0.0002751618,0.0002425905,0.0003783059],"domain_scores_gemma":[0.9990965,0.0002153581,0.00004162619,0.0004739456,0.0001146579,0.00005793146],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000001542448,0.000009847864,0.0001177622,0.00005751766,0.000008751197,3.701249e-7,0.0003436347,0.000983774,0.003500198,0.781556,0.00005860515,0.213362],"study_design_scores_gemma":[0.00019059,0.0001184702,0.00217775,0.0005558296,0.00002333208,0.0000145618,0.00002048694,0.8669259,0.03282997,0.01914206,0.07763707,0.0003639666],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.004578942,0.0002759286,0.9930724,0.0009965473,0.0001907192,0.000275938,4.114478e-7,0.000575794,0.00003332926],"genre_scores_gemma":[0.391309,0.000005743696,0.6085386,0.00005454194,0.00001374371,0.00005395761,9.705798e-8,0.000005007151,0.00001934659],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8659421,"threshold_uncertainty_score":0.3868236,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4408106796","doi":"10.22152/programming-journal.org/2025/10/8","title":"Dynamic Program Slices Change How Developers Diagnose Gradual Run-Time Type Errors","year":2025,"lang":"en","type":"article","venue":"The Art Science and Engineering of Programming","topic":"Software Engineering Research","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia; University of Alberta","funders":"","keywords":"Computer science; Programming language; Parallel computing","retraction":null,"screen_n_in":null,"score":{"opus":0.0191834616489249,"gpt":0.2784871991621939,"spread":0.2593037375132691,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001298842,0.0001571276,0.0001650983,0.0003720886,0.0002370584,0.0003468298,0.001327981,0.00003770614,7.468317e-7],"category_scores_gemma":[0.0008405802,0.0001193681,0.00003000961,0.002980416,0.000333221,0.0007247619,0.0005562123,0.0001799335,0.000007481884],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006564267,"about_ca_system_score_gemma":0.0001464414,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002661126,"about_ca_topic_score_gemma":0.000004786047,"domain_scores_codex":[0.9983952,0.00001302849,0.0001425952,0.0003604734,0.0005226471,0.0005660604],"domain_scores_gemma":[0.9989321,0.0002726522,0.00003743335,0.0004505055,0.0002026038,0.0001047673],"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":[0.000008886353,0.0001255072,0.004515984,0.0002777999,0.00006285003,0.00001852963,0.005005998,0.001118872,0.01505084,0.004626983,0.0003014588,0.9688863],"study_design_scores_gemma":[0.0006281299,0.000556452,0.1205094,0.0005928631,0.00003292474,0.00004508299,0.0003925217,0.821521,0.006954245,0.0001507894,0.04777641,0.0008402506],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9626557,0.0008836789,0.03042848,0.003549489,0.0005555754,0.0009596841,5.193029e-7,0.0009163294,0.00005050283],"genre_scores_gemma":[0.9256475,0.00004291302,0.07396314,0.00003232624,0.00001601991,0.0001297451,7.770626e-7,0.000009760911,0.0001578062],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.968046,"threshold_uncertainty_score":0.486769,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4408014589","doi":"10.22152/programming-journal.org/2026/10/8","title":"Dynamic Program Slices Change How Developers Diagnose Gradual Run-Time Type Errors","year":2025,"lang":"en","type":"article","venue":"The Art Science and Engineering of Programming","topic":"Software Engineering Research","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia; University of Alberta","funders":"","keywords":"Computer science; Type (biology); Programming language; Geology","retraction":null,"screen_n_in":null,"score":{"opus":0.0191834616489249,"gpt":0.2784871991621939,"spread":0.2593037375132691,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001298842,0.0001571276,0.0001650983,0.0003720886,0.0002370584,0.0003468298,0.001327981,0.00003770614,7.468317e-7],"category_scores_gemma":[0.0008405802,0.0001193681,0.00003000961,0.002980416,0.000333221,0.0007247619,0.0005562123,0.0001799335,0.000007481884],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006564267,"about_ca_system_score_gemma":0.0001464414,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002661126,"about_ca_topic_score_gemma":0.000004786047,"domain_scores_codex":[0.9983952,0.00001302849,0.0001425952,0.0003604734,0.0005226471,0.0005660604],"domain_scores_gemma":[0.9989321,0.0002726522,0.00003743335,0.0004505055,0.0002026038,0.0001047673],"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":[0.000008886353,0.0001255072,0.004515984,0.0002777999,0.00006285003,0.00001852963,0.005005998,0.001118872,0.01505084,0.004626983,0.0003014588,0.9688863],"study_design_scores_gemma":[0.0006281299,0.000556452,0.1205094,0.0005928631,0.00003292474,0.00004508299,0.0003925217,0.821521,0.006954245,0.0001507894,0.04777641,0.0008402506],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9626557,0.0008836789,0.03042848,0.003549489,0.0005555754,0.0009596841,5.193029e-7,0.0009163294,0.00005050283],"genre_scores_gemma":[0.9256475,0.00004291302,0.07396314,0.00003232624,0.00001601991,0.0001297451,7.770626e-7,0.000009760911,0.0001578062],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.968046,"threshold_uncertainty_score":0.486769,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}