{"id":"W7105683204","doi":"10.1109/asew67777.2025.00035","title":"ForeSPECT: A Model-Driven Framework for Validation and Traceability in Forecasting Systems","year":2025,"lang":"","type":"article","venue":"","topic":"Forecasting Techniques and Applications","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Royal Canadian Navy","funders":"NAV","keywords":"Traceability; Metamodeling; TRACE (psycholinguistics); Representation (politics); Domain (mathematical analysis); Foundation (evidence)","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":[0.02485503,0.002779234,0.001968309,0.005375895,0.002261789,0.008759425,0.007083297,0.003479557,0.003670348],"category_scores_gemma":[0.06137406,0.002435653,0.007274521,0.003235194,0.003552965,0.009226472,0.008112041,0.007147647,0.001485969],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003821554,"about_ca_system_score_gemma":0.0124027,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03644373,"about_ca_topic_score_gemma":0.03961569,"domain_scores_codex":[0.9837916,0.006420772,0.002397418,0.002078025,0.004596537,0.0007157088],"domain_scores_gemma":[0.9661167,0.01747729,0.002506108,0.008609926,0.004576938,0.0007129613],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002564159,0.0003591138,0.005660164,0.001142058,0.0006014821,0.0009877292,0.001454054,0.5807868,0.004634109,0.2453694,0.01456116,0.1441875],"study_design_scores_gemma":[0.00004816451,0.00006418692,0.0003321238,0.0003814786,0.00009742748,0.0001644006,0.0001745858,0.8291709,0.00465099,0.1345634,0.03025173,0.0001006064],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0008222804,0.00009953161,0.9913433,0.0003200149,0.00003740339,0.0001970818,0.0005640439,0.006029883,0.0005864502],"genre_scores_gemma":[0.03895072,0.0003718929,0.9531385,0.0003010004,0.00005537372,0.0007032949,0.004237821,0.001521713,0.0007196789],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.03644373,"threshold_uncertainty_score":0.1314476,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2125855422147073,"score_gpt":0.4343277666536215,"score_spread":0.2217422244389142,"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."}}