{"id":"W2889038510","doi":"10.3389/feduc.2018.00068","title":"Validity Evidence for Progress Monitoring With Star Reading: Slope Estimates, Administration Frequency, and Number of Data Points","year":2018,"lang":"en","type":"article","venue":"Frontiers in Education","topic":"Reading and Literacy Development","field":"Psychology","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Reading (process); Star (game theory); Data collection; Computer science; Outlier; Estimator; Psychology; Statistics; Mathematics; Artificial intelligence","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004322713,0.0000930539,0.0001307153,0.00007337138,0.00005471005,0.00003553879,0.0001809785,0.00005551022,0.00001995905],"category_scores_gemma":[0.0001424792,0.00008605137,0.000007088341,0.000135489,0.0001093708,0.0004424012,0.00002126914,0.00005699155,0.000002810146],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006270547,"about_ca_system_score_gemma":0.0002492091,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009499175,"about_ca_topic_score_gemma":0.000005456524,"domain_scores_codex":[0.9991275,0.00004958053,0.0002336429,0.0003350011,0.0001035504,0.0001507393],"domain_scores_gemma":[0.9992272,0.00007942541,0.0001523988,0.0003875104,0.0001088904,0.00004453943],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001667135,0.0001307317,0.9448537,0.00006942718,0.00001902514,4.6905e-7,0.003800862,3.282269e-8,0.0000342217,0.0002185095,0.006024356,0.04468193],"study_design_scores_gemma":[0.0004569728,0.0003454247,0.9867833,0.001043736,0.00005433121,0.00002666569,0.002823554,0.0001541268,0.002228947,0.002671644,0.003174905,0.0002363862],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9882576,0.000496619,0.006962327,0.0002674195,0.002824407,0.0004558472,0.00001611982,0.00001968884,0.0006999714],"genre_scores_gemma":[0.6480018,0.0000180681,0.3515091,0.00000826767,0.0001615533,0.00008392289,0.00007018574,0.000008765928,0.000138308],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3445468,"threshold_uncertainty_score":0.3509073,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09735874007776728,"score_gpt":0.4260005749852975,"score_spread":0.3286418349075302,"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."}}