{"id":"W2112673666","doi":"","title":"Understanding the Connections between Large-Scale Assessment and School Improvement Planning.","year":2010,"lang":"en","type":"article","venue":"Canadian Journal of Educational Administration and Policy","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Scale (ratio); Professional development; Psychology; Medical education; Pedagogy; Protocol (science); Mathematics education; Medicine; Geography","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001462582,0.0001046049,0.0001343083,0.0004315602,0.0007423387,0.0006479958,0.0002316408,0.00005295386,0.00113326],"category_scores_gemma":[0.000990068,0.00007251033,0.000047693,0.0003033231,0.0001638318,0.0003462731,0.00001587631,0.0003273536,0.000008744348],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002159677,"about_ca_system_score_gemma":0.01057748,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003320017,"about_ca_topic_score_gemma":0.007233827,"domain_scores_codex":[0.9986055,0.00005888254,0.0005429253,0.0001603319,0.0004287678,0.0002035707],"domain_scores_gemma":[0.9973941,0.0009968837,0.0003638388,0.0001525524,0.0003410172,0.0007516386],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000005105636,0.0000519577,0.6248768,0.000003804157,0.0000416416,6.849662e-7,0.001322078,0.000004420142,0.0003451753,0.3460702,0.02613064,0.001147522],"study_design_scores_gemma":[0.0002596792,0.0002078235,0.813089,0.00001101427,0.00002140269,0.00005898338,0.008646543,0.00005135258,0.00003533146,0.1271636,0.05035483,0.0001003856],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7882684,0.000058258,0.001385134,0.2024733,0.000969358,0.0001683247,0.00007053353,0.000001770127,0.006604926],"genre_scores_gemma":[0.9946538,0.000005464257,0.0006535926,0.001360843,0.001616442,0.000008505634,0.00001427287,0.000005507738,0.001681631],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2189066,"threshold_uncertainty_score":0.9997798,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1532314622462823,"score_gpt":0.4521814844311864,"score_spread":0.2989500221849041,"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."}}