{"id":"W4407986476","doi":"10.1007/s10664-025-10618-0","title":"Evaluating interactive documentation for programmers","year":2025,"lang":"en","type":"article","venue":"Empirical Software Engineering","topic":"Software Engineering Research","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":false,"ca_institutions":"McGill University","funders":"Fonds de recherche du Québec – Nature et technologies; Natural Sciences and Engineering Research Council of Canada","keywords":"Documentation; Computer science; World Wide Web; Software engineering; Programming language","routes":{"ca_aff":true,"ca_fund":true,"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":[],"consensus_categories":[],"category_scores_codex":[0.0004843961,0.0001826722,0.0001809712,0.0002806352,0.00009707376,0.0002444249,0.0005816056,0.00007684704,0.000008679946],"category_scores_gemma":[0.006459929,0.000188949,0.00008802665,0.0008658777,0.0000140786,0.0005133756,0.0002286604,0.000239977,0.0000155374],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002938457,"about_ca_system_score_gemma":0.0001029484,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000007656922,"about_ca_topic_score_gemma":4.583866e-7,"domain_scores_codex":[0.9984918,0.00002611233,0.0002583872,0.0004496453,0.0003059309,0.0004681645],"domain_scores_gemma":[0.9961793,0.00316117,0.00003588585,0.0003565926,0.0001650233,0.0001020006],"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":[0.0001010268,0.0003225047,0.1560905,0.001055956,0.0005492997,0.00002635933,0.002637753,0.220659,0.004119759,0.00587531,0.01141683,0.5971457],"study_design_scores_gemma":[0.002311099,0.000652872,0.1217766,0.0004496665,0.00004446544,0.0000129177,0.000051832,0.8232046,0.01098133,0.002149849,0.03734136,0.001023424],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0462276,0.0001325639,0.9506574,0.0004972745,0.0007633619,0.0005462337,0.00000174382,0.00115608,0.00001775961],"genre_scores_gemma":[0.4056336,0.000002342962,0.5931816,0.0001921536,0.00008881032,0.0005793911,0.00001074406,0.0000273067,0.000284084],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6025456,"threshold_uncertainty_score":0.7733605,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04750618481790624,"score_gpt":0.4132381797684386,"score_spread":0.3657319949505324,"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."}}