{"id":"W6949396649","doi":"10.5281/zenodo.14714469","title":"ICST 2025 UAV Testing Competition: Evaluation Artifacts","year":2025,"lang":"en","type":"dataset","venue":"PolyPublie (École Polytechnique de Montréal)","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Ranking (information retrieval); Software; Test (biology); Competition (biology); Software testing","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":[0.006547346,0.005333665,0.001721265,0.00573812,0.001560185,0.003441963,0.005561543,0.003516083,0.01935899],"category_scores_gemma":[0.02466912,0.00115117,0.002319113,0.006183776,0.0009440891,0.002138814,0.003909409,0.003081671,0.02880507],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003811929,"about_ca_system_score_gemma":0.00417929,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.04585187,"about_ca_topic_score_gemma":0.08987423,"domain_scores_codex":[0.9899744,0.002949626,0.001294778,0.001444995,0.003655266,0.0006808917],"domain_scores_gemma":[0.9847181,0.00462857,0.001006192,0.003710022,0.004971182,0.0009658472],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002037622,0.0002046073,0.001545497,0.0007142919,0.00007561356,0.000108036,0.00004166616,0.001773279,0.0004713706,0.0005621393,0.9842299,0.01006985],"study_design_scores_gemma":[0.001751943,0.0004699118,0.0209298,0.000839823,0.0001864558,0.0007948574,0.0002815117,0.02420576,0.006260525,0.004276879,0.9397898,0.0002127635],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.006110415,0.0005273807,0.002797053,0.0005776803,0.0003103148,0.0004740112,0.9751807,0.00854363,0.005478673],"genre_scores_gemma":[0.002235085,0.0000636891,0.002283659,0.0001162223,0.0000131688,0.0003314329,0.9936678,0.0003136942,0.0009752795],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.04585187,"threshold_uncertainty_score":0.09116995,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03074036814795914,"score_gpt":0.2891085089319124,"score_spread":0.2583681407839532,"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."}}