{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":2,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":2,"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","author_layer_release":"2026-06-26"},"query_hash":"252a22f28e0a","filters":{"venue":"IEEE International Software Metrics Symposium"}},"results":[{"id":"W1554594814","doi":"10.1109/metrics.2004.2","title":"A prototype empirical evaluation of test driven development","year":2004,"lang":"en","type":"article","venue":"IEEE International Software Metrics Symposium","topic":"Software Engineering Research","field":"Computer Science","cited_by":65,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta; University of Calgary","funders":"","keywords":"Computer science; Debugging; Test-driven development; Software development; Software development process; Software engineering; Test strategy; Software quality; New product development; Manual testing; Test (biology); Software; Empirical research; Software construction; Operating system; Business","authors":[{"name":"Adam Geras","is_ca":true},{"name":"Michael R. Smith","is_ca":true},{"name":"James Miller","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.06676157119916065,"gpt":0.3508707110518867,"spread":0.284109139852726,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001228565,0.0001647953,0.0001708713,0.0008969278,0.00006403838,0.0001080886,0.001431434,0.00009695873,0.00002008164],"category_scores_gemma":[0.005648719,0.000163685,0.00007635322,0.001581987,0.00003940619,0.0003723094,0.0002392792,0.0002007738,0.0000943953],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000985409,"about_ca_system_score_gemma":0.0007818536,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002737584,"about_ca_topic_score_gemma":0.000005193571,"domain_scores_codex":[0.9961749,0.00004834132,0.0004349757,0.0004233662,0.002643997,0.0002744387],"domain_scores_gemma":[0.9967852,0.001015923,0.0001392891,0.0004136915,0.001524754,0.0001211965],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000400811,0.001835106,0.7417122,0.0001085084,0.0003442112,0.00005155062,0.002943401,0.1749297,0.008195251,0.001797787,0.001003745,0.06703844],"study_design_scores_gemma":[0.007937169,0.001326284,0.5419647,0.0004983828,0.00009320378,0.0001787733,0.00002776621,0.1191941,0.2994325,0.008214883,0.01926709,0.001865123],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1465522,0.00006476886,0.8501034,0.0007304325,0.001389169,0.0006361086,0.000007512615,0.0002903198,0.0002261247],"genre_scores_gemma":[0.8277234,0.000008063669,0.1718519,0.00005109359,0.00009626178,0.0001837218,0.0000101819,0.00001735744,0.00005796095],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6811713,"threshold_uncertainty_score":0.6762452,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1532055694","doi":"10.1109/metrics.2004.18","title":"COTS acquisition process: incorporating business factors in COTS vendor evaluation taxonomy","year":2004,"lang":"en","type":"article","venue":"IEEE International Software Metrics Symposium","topic":"Technology Assessment and Management","field":"Engineering","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 Alberta","funders":"","keywords":"Vendor; Process (computing); Computer science; Product (mathematics); Business process; Process management; Risk analysis (engineering); Business; Marketing; Work in process; Operating system","authors":[{"name":"Hean Chin Yeoh","is_ca":true},{"name":"James Miller","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02804425957723891,"gpt":0.2732348847340538,"spread":0.2451906251568149,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003376914,0.0002096172,0.0001773796,0.001167366,0.00005754748,0.00007859104,0.0003421593,0.000148784,0.0000293896],"category_scores_gemma":[0.0001672725,0.0002297978,0.00004362552,0.001408112,0.00003171809,0.0004402375,0.00004434659,0.0002058554,0.00002258425],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001043935,"about_ca_system_score_gemma":0.00006567986,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005373119,"about_ca_topic_score_gemma":0.00006769117,"domain_scores_codex":[0.9984931,0.00001924479,0.000384992,0.0002721347,0.000608441,0.0002221109],"domain_scores_gemma":[0.9992505,0.00007453027,0.0001359209,0.0001603755,0.0003380224,0.00004069975],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.00001681917,0.0002421972,0.1936616,0.0001261592,0.0001424727,0.00002821038,0.0002037554,0.7923481,0.003502986,0.001421199,0.0002447557,0.008061725],"study_design_scores_gemma":[0.01679631,0.0004121879,0.6209604,0.001330135,0.0004588981,0.00005525195,0.001177588,0.1351438,0.1571746,0.05416495,0.008395066,0.00393085],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7542849,0.00009928019,0.2396495,0.0002795726,0.003056158,0.0006975311,0.00002581852,0.0006360129,0.001271188],"genre_scores_gemma":[0.9951633,0.00003128181,0.004053624,0.00004096117,0.0001227837,0.0003313268,0.0002026695,0.00003226586,0.00002174331],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6572043,"threshold_uncertainty_score":0.9370881,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}