{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":6,"total_is_capped":false,"direct_labels_cover":1,"predictions_cover":6,"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":"cba7bb08c2d0","filters":{"venue":"25th International Conference on Software Engineering, 2003. Proceedings."}},"results":[{"id":"W4241947741","doi":"10.1109/icse.2003.1201219","title":"Hipikat: recommending pertinent software development artifacts","year":2003,"lang":"en","type":"article","venue":"25th International Conference on Software Engineering, 2003. Proceedings.","topic":"Software Engineering Research","field":"Computer Science","cited_by":173,"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","funders":"U.S. Consumer Product Safety Commission","keywords":"Eclipse; Computer science; Task (project management); Context (archaeology); Open source; Open source software; Software; Software project management; Open-source software development; Software engineering; Software development; World Wide Web; Data science; Software construction; Engineering; Systems engineering; Programming language","retraction":null,"screen_n_in":null,"score":{"opus":0.04131285254867295,"gpt":0.2738731626548143,"spread":0.2325603101061413,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0007418304,0.0006179772,0.0003940021,0.0007035332,0.0002148184,0.0007232466,0.00186843,0.0002158052,0.0005021125],"category_scores_gemma":[0.008888165,0.0006409551,0.000100401,0.001115518,0.00004021115,0.0009278269,0.0003495677,0.0008012069,0.0004886297],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007236084,"about_ca_system_score_gemma":0.0004335513,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00000465133,"about_ca_topic_score_gemma":0.000001374109,"domain_scores_codex":[0.9958242,0.00001956364,0.0006819388,0.001059481,0.001392533,0.001022351],"domain_scores_gemma":[0.9974015,0.00003699644,0.0001986294,0.0004398217,0.001447542,0.0004754853],"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":[0.00001828534,0.0002186443,0.02652894,0.0001220714,0.0001621846,0.00005819723,0.0009845519,0.0009985984,0.0001485009,0.9609392,0.00788668,0.001934175],"study_design_scores_gemma":[0.002588005,0.0005610504,0.03549812,0.001544763,0.00002764361,0.0003294319,0.000232706,0.4884448,0.05585932,0.0002611228,0.4107271,0.003925951],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05078409,0.0001074278,0.9399742,0.00008664039,0.003504812,0.0005595299,0.00001130667,0.002662751,0.002309217],"genre_scores_gemma":[0.1843691,0.00004355786,0.8131911,0.0002751253,0.0001752264,0.000248815,0.00002525957,0.00009524102,0.001576531],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.960678,"threshold_uncertainty_score":0.9996042,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4229596932","doi":"10.1109/icse.2003.1201189","title":"Using benchmarking to advance research: a challenge to software engineering","year":2003,"lang":"en","type":"article","venue":"25th International Conference on Software Engineering, 2003. Proceedings.","topic":"Software Engineering Research","field":"Computer Science","cited_by":150,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Benchmarking; Benchmark (surveying); Computer science; Data science; Software engineering; Software; Academic community; Engineering management; Management science; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.1101268605763671,"gpt":0.3553565504832152,"spread":0.2452296899068481,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001606062,0.0007129067,0.0005186179,0.00200932,0.0002552777,0.00086891,0.00282113,0.0002609519,0.0001812547],"category_scores_gemma":[0.02405716,0.0008155166,0.0001168275,0.003684172,0.00004172446,0.001062824,0.0008015716,0.001321366,0.000329014],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001114423,"about_ca_system_score_gemma":0.0003632354,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002198966,"about_ca_topic_score_gemma":0.00000306171,"domain_scores_codex":[0.9935786,0.00003417471,0.0006685107,0.001602658,0.00239492,0.001721186],"domain_scores_gemma":[0.9948,0.00009815988,0.000114142,0.0007420091,0.00324367,0.001001977],"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":[0.00002660777,0.0001408052,0.003339393,0.0001527551,0.00009051163,0.00009192485,0.001150655,0.04131754,0.0005259609,0.9495891,0.002829731,0.000744961],"study_design_scores_gemma":[0.0008193678,0.0008566037,0.004067846,0.001747629,0.00001117939,0.0001344538,0.0001256912,0.8810512,0.006227043,0.000123015,0.1028016,0.002034442],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03554428,0.0001257,0.9582456,0.00007472648,0.002669547,0.0009707446,0.00002500254,0.001614498,0.0007298615],"genre_scores_gemma":[0.1309175,0.00003994091,0.8674935,0.0002010498,0.0004453416,0.0003523687,0.000007128916,0.0001439733,0.0003992042],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9494662,"threshold_uncertainty_score":0.9994296,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4238417811","doi":"10.1109/icse.2003.1201304","title":"FEAT a tool for locating, describing, and analyzing concerns in source code","year":2003,"lang":"en","type":"article","venue":"25th International Conference on Software Engineering, 2003. Proceedings.","topic":"Software Engineering Research","field":"Computer Science","cited_by":33,"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","funders":"","keywords":"Computer science; Source code; Task (project management); Code (set theory); Programming language; Software engineering; Code review; KPI-driven code analysis; Open source; Static program analysis; Software development; Software; Systems engineering; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.05235386629265224,"gpt":0.2912926794672911,"spread":0.2389388131746389,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0006974671,0.0003849393,0.0003274881,0.0005471408,0.00009816846,0.0005527484,0.000945474,0.0001576494,0.00004420677],"category_scores_gemma":[0.007406188,0.0004197047,0.00006075957,0.0008141478,0.00005296658,0.0006431326,0.0001742102,0.0004755307,0.00001585317],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003187042,"about_ca_system_score_gemma":0.0001955814,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001898122,"about_ca_topic_score_gemma":0.000005691305,"domain_scores_codex":[0.9974908,0.00001279467,0.0004429999,0.0008156144,0.0005694883,0.0006683251],"domain_scores_gemma":[0.9984379,0.00005554315,0.0001291092,0.000248098,0.0009366769,0.0001926055],"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":[0.00001666475,0.00006239725,0.0761839,0.0001056124,0.00005251048,0.00001120624,0.0006518177,0.002017005,0.0001319003,0.9184999,0.001486073,0.0007809982],"study_design_scores_gemma":[0.00123222,0.0002043071,0.005111973,0.0004853042,0.000008257255,0.00004293039,0.00008644776,0.9747053,0.002783743,0.00010287,0.01452743,0.0007091663],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.09771533,0.0001439312,0.9000428,0.00003027491,0.0005936649,0.0005330077,0.00001677911,0.0006027492,0.0003214109],"genre_scores_gemma":[0.4997269,0.00006298437,0.4986689,0.0001191,0.0001393002,0.000294392,0.00001063352,0.00006574851,0.0009119841],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9726883,"threshold_uncertainty_score":0.9998255,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4239921431","doi":"10.1109/icse.2003.1201214","title":"Design pattern rationale graphs: linking design to source","year":2003,"lang":"en","type":"article","venue":"25th International Conference on Software Engineering, 2003. Proceedings.","topic":"Software Engineering Research","field":"Computer Science","cited_by":19,"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","funders":"","keywords":"Software design pattern; Computer science; Design pattern; Structural pattern; Flexibility (engineering); Pattern language (formal languages); Source code; TRACE (psycholinguistics); Code (set theory); Representation (politics); Software engineering; Human–computer interaction; Programming language; Software design; Software development; Software","retraction":null,"screen_n_in":null,"score":{"opus":0.05281837649059293,"gpt":0.2709705174956941,"spread":0.2181521410051012,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001032416,0.0005130938,0.0003248473,0.000830862,0.00015932,0.0008128466,0.001937865,0.0001896698,0.000206569],"category_scores_gemma":[0.005423454,0.0005502006,0.00008401484,0.001486561,0.00003317575,0.0007606384,0.0002271338,0.0006161601,0.0003672117],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003216682,"about_ca_system_score_gemma":0.0003153906,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000006390514,"about_ca_topic_score_gemma":4.271043e-7,"domain_scores_codex":[0.9963248,0.00004037441,0.0004970335,0.0009774066,0.001367687,0.0007927357],"domain_scores_gemma":[0.9972646,0.00007659861,0.0001352796,0.0004023052,0.001704569,0.0004167086],"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":[0.00002914521,0.000160164,0.006726374,0.00006408311,0.000146626,0.00004480292,0.001276527,0.06788725,0.0004164122,0.9120719,0.009047709,0.002129005],"study_design_scores_gemma":[0.0006502035,0.0003646596,0.001638855,0.0003602357,0.00000872476,0.00007204171,0.00004165695,0.9687303,0.01415829,0.0002606738,0.01276287,0.0009515283],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002218448,0.00004707012,0.9940041,0.00005326737,0.001296378,0.0007682221,0.000005941551,0.001265169,0.000341357],"genre_scores_gemma":[0.1276311,0.00002723701,0.8701012,0.0005073466,0.0001587195,0.0003973827,0.00001095738,0.00008900459,0.001077043],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9118112,"threshold_uncertainty_score":0.9996949,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4237741047","doi":"10.1109/icse.2003.1201251","title":"Empowering software engineers in human-centered design","year":2003,"lang":"en","type":"article","venue":"25th International Conference on Software Engineering, 2003. Proceedings.","topic":"Usability and User Interface Design","field":"Computer Science","cited_by":15,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Concordia University","funders":"","keywords":"Usability; Computer science; Software engineering; Usability engineering; Social software engineering; Software development; Toolbox; Software construction; Human–computer interaction; Software; Programming language","retraction":null,"screen_n_in":null,"score":{"opus":0.05692146588363678,"gpt":0.2961389041541138,"spread":0.239217438270477,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0006508565,0.0004992296,0.0003741671,0.0005598503,0.0000957845,0.0004839907,0.001599531,0.0002106216,0.0002717176],"category_scores_gemma":[0.002067816,0.0005445103,0.00009289448,0.0008699322,0.00004488129,0.00102558,0.0001519857,0.0006363749,0.00009737183],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003985298,"about_ca_system_score_gemma":0.0001649962,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001191105,"about_ca_topic_score_gemma":0.000002837292,"domain_scores_codex":[0.9971233,0.00003031914,0.0006058166,0.000879981,0.0006862266,0.0006743293],"domain_scores_gemma":[0.998455,0.00001460947,0.0001584298,0.0003563054,0.0007957551,0.0002199313],"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":[0.00002348691,0.0002198343,0.004196592,0.00006191764,0.00006023084,0.00003893202,0.001131738,0.006549083,0.0004041943,0.984999,0.002195352,0.0001196505],"study_design_scores_gemma":[0.002083796,0.0005889924,0.003219277,0.0008160347,0.00001453012,0.00008137467,0.0002582008,0.944106,0.03419318,0.0007933844,0.01213604,0.001709129],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0261269,0.00005059004,0.9696379,0.0000267152,0.001469393,0.0004511718,0.0000101707,0.000829771,0.00139743],"genre_scores_gemma":[0.4134693,0.00002332063,0.5854827,0.000182256,0.00006872037,0.0001338864,0.000009124604,0.00005330751,0.000577415],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9842056,"threshold_uncertainty_score":0.9997007,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4243475163","doi":"10.1109/icse.2003.1201295","title":"/spl chi/Chek: A model checker for multi-valued reasoning","year":2003,"lang":"en","type":"article","venue":"25th International Conference on Software Engineering, 2003. Proceedings.","topic":"Formal Methods in Verification","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Model checking; Symbolic trajectory evaluation; Computer science; Computation tree logic; Generalization; Extension (predicate logic); Temporal logic; Theoretical computer science; Programming language; Linear temporal logic; Algorithm; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.08485247251597294,"gpt":0.3296170011743911,"spread":0.2447645286584182,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0007249785,0.0003692214,0.0002673024,0.0002740998,0.000130982,0.0003534368,0.001176819,0.0001835368,0.0000433712],"category_scores_gemma":[0.00381645,0.0003846654,0.00009774892,0.0005100978,0.00003771175,0.0009235076,0.0001095124,0.0003618209,0.00004177923],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002369545,"about_ca_system_score_gemma":0.0002084768,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000003135176,"about_ca_topic_score_gemma":4.397157e-7,"domain_scores_codex":[0.9977755,0.00001223579,0.0004427399,0.0007112952,0.0005614063,0.0004968052],"domain_scores_gemma":[0.9978247,5.444383e-7,0.0002419575,0.0002921201,0.001462086,0.0001785375],"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":[0.00001444371,0.00007265215,0.0002650342,0.00003880902,0.0000362389,0.000001124645,0.0003574707,0.002266979,0.00002670595,0.9962275,0.0006625518,0.00003048333],"study_design_scores_gemma":[0.0006499531,0.00008610203,0.0002970695,0.00012325,0.000008888057,0.00001791015,0.00003151321,0.9879688,0.006716176,0.00007106805,0.003603867,0.0004253922],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002718195,0.00003157528,0.9922743,0.000001675859,0.00111588,0.0004789735,0.00002116854,0.0006091866,0.00274908],"genre_scores_gemma":[0.003237636,0.0000265289,0.9946276,0.0002111434,0.0001031799,0.0003336993,0.00001649073,0.00004991551,0.001393821],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9961565,"threshold_uncertainty_score":0.9998605,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"}]}