{"meta":{"query_hash":"94626412b82e","filters":{"venue":"Educational Assessment"},"cohort_total":6,"direct_labels_cover":0,"predictions_cover":6,"exported":6,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/94626412b82e","api":"https://metacan.xera.ac/api/v1/cohort?venue=Educational+Assessment"},"results":[{"id":"W1999929076","doi":"10.1080/10627197.2011.584042","title":"Voices From Test-Takers: Further Evidence for Language Assessment Validation and Use","year":2011,"lang":"en","type":"article","venue":"Educational Assessment","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":56,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Guangdong University of Foreign Studies; Ministry of Education, India; Ministry of Earth Sciences","keywords":"Test (biology); Language assessment; Psychology; Test validity; Scale (ratio); Coding (social sciences); Test score; Computer science; Psychometrics; Mathematics education; Standardized test; Developmental psychology","score_opus":0.14416238535195033,"score_gpt":0.4552257976525222,"score_spread":0.3110634123005719,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1999929076","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97812563,0.0029714415,0.0061020222,0.007367013,0.00018267517,0.00011528806,0.00010727189,0.000046955684,0.004981703],"genre_scores_gemma":[0.99455243,0.0007247289,0.0018305904,0.001846136,0.000110603796,0.00012430765,0.00008408133,0.00006191035,0.00066523167],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.6754895,0.23173456,0.02039349,0.008941623,0.05831548,0.0051253242],"domain_scores_gemma":[0.29655135,0.58150554,0.04345421,0.025133682,0.047002777,0.006352395],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.16795583,0.0006862718,0.001460696,0.003745166,0.0050467476,0.007357724,0.0030838372,0.0028186003,0.0026675428],"category_scores_gemma":[0.5089094,0.0008900915,0.001389184,0.0020104637,0.009811517,0.00712669,0.01106563,0.0049013714,0.00060939963],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00034389884,0.00013000949,0.078418285,0.0005830973,0.00011875083,0.000871143,0.8657131,0.000051525996,0.0025173414,0.0010073405,0.00075333146,0.049492124],"study_design_scores_gemma":[0.00011419213,0.0011644597,0.14296308,0.0027134626,0.00016570723,0.0031433324,0.8113228,0.0010015107,0.0071906666,0.0033458078,0.026594765,0.00028021532],"about_ca_topic_score_codex":0.004104786,"about_ca_topic_score_gemma":0.0037134096,"teacher_disagreement_score":0.16795583,"about_ca_system_score_codex":0.002931892,"about_ca_system_score_gemma":0.0038026855,"threshold_uncertainty_score":0.88824594},"labels":[],"label_agreement":null},{"id":"W2072296938","doi":"10.1080/10627190802394222","title":"Validity Issues in Assessing English Language Learners' Language Proficiency","year":2008,"lang":"en","type":"article","venue":"Educational Assessment","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":47,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Memorial University of Newfoundland","keywords":"Accountability; Legislation; Language proficiency; Language assessment; Psychology; English language; Pedagogy; Medical education; Mathematics education; Political science; Medicine","score_opus":0.06769468706590388,"score_gpt":0.4642791324549266,"score_spread":0.3965844453890227,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2072296938","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.21058784,0.025953574,0.42742273,0.13049065,0.005109836,0.008753577,0.0009393753,0.00043304538,0.19030945],"genre_scores_gemma":[0.7472276,0.0043812855,0.22412477,0.011835191,0.0012615166,0.0073690424,0.00045294952,0.00024166756,0.003105988],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.4212688,0.42171198,0.049436353,0.012755522,0.09128568,0.0035417308],"domain_scores_gemma":[0.23934801,0.6323883,0.025099477,0.025532117,0.07563836,0.0019937796],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.47553632,0.0007453884,0.001628255,0.0070863827,0.0068432237,0.008754864,0.0032770468,0.002809954,0.0015335141],"category_scores_gemma":[0.6634406,0.00088358915,0.0019596703,0.0060584177,0.021868713,0.009952812,0.008223538,0.004860862,0.00056310074],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00042945665,0.0004156962,0.21847975,0.0050362675,0.000792253,0.00048079138,0.057277065,0.0023460565,0.0017108291,0.30535558,0.012098955,0.39557728],"study_design_scores_gemma":[0.00034797832,0.0017416596,0.21084099,0.022884049,0.0009086266,0.001807904,0.06677267,0.020334044,0.01107296,0.5300572,0.13277441,0.00045752883],"about_ca_topic_score_codex":0.011278531,"about_ca_topic_score_gemma":0.0119093815,"teacher_disagreement_score":0.47553632,"about_ca_system_score_codex":0.00895578,"about_ca_system_score_gemma":0.02553042,"threshold_uncertainty_score":0.6467571},"labels":[],"label_agreement":null},{"id":"W2086203028","doi":"10.1080/10627190802602384","title":"Identifying Potential Test Item Misalignment Using Student Verbal Reports","year":2008,"lang":"en","type":"article","venue":"Educational Assessment","topic":"Educational Strategies and Epistemologies","field":"Psychology","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; University of Alberta","funders":"","keywords":"Ambiguity; Psychology; Cognition; Construct (python library); Test (biology); Item analysis; Confidence interval; Cognitive psychology; Applied psychology; Psychometrics; Statistics; Developmental psychology; Computer science; Mathematics","score_opus":0.08693562939927937,"score_gpt":0.43889796600229253,"score_spread":0.3519623366030131,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2086203028","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9472997,0.00037071286,0.04944824,0.00014771773,0.00007271369,0.000431522,0.00020377505,0.00019308476,0.001832506],"genre_scores_gemma":[0.9745072,0.00015534427,0.024158485,0.00006773815,0.00003572717,0.00044208413,0.0003155747,0.000034922326,0.0002829187],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.93126684,0.035412494,0.014325989,0.002643793,0.015516278,0.00083448994],"domain_scores_gemma":[0.5858323,0.29395482,0.07054771,0.021230107,0.02727257,0.0011624694],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.054042775,0.0009933667,0.0011544565,0.0043457025,0.00045726885,0.0030078916,0.0010432394,0.00078940473,0.0010020402],"category_scores_gemma":[0.32486087,0.00049428135,0.0006451857,0.0039795437,0.0009466823,0.0023742465,0.0020740544,0.0010840168,0.00035258953],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013382682,0.00045833256,0.7828666,0.0006248938,0.00041592558,0.00069083506,0.0108988825,0.0023493646,0.0074411756,0.001720978,0.0007234465,0.19047129],"study_design_scores_gemma":[0.00015346559,0.004153951,0.87402636,0.0009243317,0.00048465776,0.0032186792,0.01457736,0.046421003,0.039948612,0.009946895,0.0058237193,0.00032084002],"about_ca_topic_score_codex":0.0004899047,"about_ca_topic_score_gemma":0.0005945046,"teacher_disagreement_score":0.054042775,"about_ca_system_score_codex":0.0006401995,"about_ca_system_score_gemma":0.00085597124,"threshold_uncertainty_score":0.28580892},"labels":[],"label_agreement":null},{"id":"W2523280788","doi":"10.1080/10627197.2016.1236677","title":"Approaches to Classroom Assessment Inventory: A New Instrument to Support Teacher Assessment Literacy","year":2016,"lang":"en","type":"article","venue":"Educational Assessment","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":112,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"","keywords":"Accountability; Literacy; Construct (python library); Educational assessment; Psychology; Construct validity; Mathematics education; Standards for Educational and Psychological Testing; Standardized test; Standards-based assessment; Process (computing); Authentic assessment; Pedagogy; Psychometrics; Higher education; Computer science; Political science; Education theory; Curriculum","score_opus":0.1328510477980371,"score_gpt":0.4138136676963231,"score_spread":0.28096261989828597,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2523280788","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.51862025,0.0032803665,0.30645108,0.008135222,0.0012549064,0.0142283365,0.010333793,0.009342569,0.12835348],"genre_scores_gemma":[0.34230158,0.0016432806,0.625768,0.0009850942,0.00025262713,0.013551587,0.003683569,0.00056000095,0.011254266],"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.99148566,0.003093509,0.0015499623,0.000520675,0.00306019,0.00029003498],"domain_scores_gemma":[0.97121584,0.01304952,0.004314045,0.0020226561,0.007819675,0.0015782956],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008253588,0.00043044097,0.0005822259,0.0038586024,0.0009327062,0.0019965426,0.0010598757,0.00040752764,0.0030607965],"category_scores_gemma":[0.034305673,0.00044070103,0.0006528134,0.002093418,0.0008376984,0.0031293628,0.0036657213,0.0025388037,0.001466707],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016975823,0.0012260206,0.16603029,0.00044914393,0.00008644386,0.00019090564,0.011236252,0.00089482113,0.004672242,0.0062153847,0.027115552,0.7817132],"study_design_scores_gemma":[0.00026054282,0.0017751808,0.68682206,0.0009772708,0.00022402017,0.0018730275,0.010905188,0.011249821,0.006550833,0.016683644,0.26232648,0.00035193362],"about_ca_topic_score_codex":0.0034212815,"about_ca_topic_score_gemma":0.009186864,"teacher_disagreement_score":0.008253588,"about_ca_system_score_codex":0.001490116,"about_ca_system_score_gemma":0.0059663807,"threshold_uncertainty_score":0.043649673},"labels":[],"label_agreement":null},{"id":"W2975545495","doi":"10.1080/10627197.2019.1670056","title":"Toward a Teacher Professional Learning Continuum in Assessment for Learning","year":2019,"lang":"en","type":"article","venue":"Educational Assessment","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":52,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Council of Ministers of Education; Queen's University","funders":"","keywords":"Mathematics education; Assessment for learning; Professional learning community; Psychology; Pedagogy; Professional development; Observational study; Empirical research; Epistemology; Mathematics; Formative assessment","score_opus":0.033801063854011035,"score_gpt":0.43057365010553394,"score_spread":0.3967725862515229,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2975545495","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.87484074,0.00051947124,0.06165882,0.012097061,0.000076058735,0.00040146758,0.00008001733,0.00037850227,0.049947884],"genre_scores_gemma":[0.9613699,0.0001060921,0.035403416,0.0004674648,0.000012141065,0.00022968161,0.000037454396,0.00003789091,0.0023359794],"study_design_codex":"qualitative","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9834599,0.009704266,0.0011398343,0.0012342974,0.0035096314,0.00095216656],"domain_scores_gemma":[0.96588945,0.015916003,0.00351245,0.0038391289,0.006361803,0.0044812365],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017388906,0.00016276124,0.0002504753,0.0017457985,0.0040564546,0.008004634,0.0011553685,0.0016953044,0.0016640059],"category_scores_gemma":[0.032350212,0.0005196913,0.0002965005,0.0011026826,0.007727739,0.006788302,0.008038673,0.0043521896,0.00073528057],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00018791339,0.0013222891,0.117604285,0.0004200417,0.000014224592,0.0017761218,0.5055659,0.001008339,0.0114137065,0.1326165,0.0031841742,0.22488657],"study_design_scores_gemma":[0.000109951194,0.0010138722,0.15712236,0.00081468705,0.000009832429,0.004280857,0.49639496,0.0070038624,0.007870055,0.19565524,0.12954569,0.00017868895],"about_ca_topic_score_codex":0.0020043594,"about_ca_topic_score_gemma":0.0036081276,"teacher_disagreement_score":0.017388906,"about_ca_system_score_codex":0.0050151953,"about_ca_system_score_gemma":0.010427457,"threshold_uncertainty_score":0.09196246},"labels":[],"label_agreement":null},{"id":"W3113373418","doi":"10.1080/10627197.2020.1858783","title":"The Effect of Linguistic Factors on Assessment of English Language Learners’ Mathematical Ability: A Differential Item Functioning Analysis","year":2020,"lang":"en","type":"article","venue":"Educational Assessment","topic":"Cognitive and developmental aspects of mathematical skills","field":"Mathematics","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Differential item functioning; Ell; Psychology; Achievement test; Test (biology); Standardized test; Mathematics education; Language proficiency; Item response theory; Confirmatory factor analysis; Item analysis; Language assessment; Linguistics; Psychometrics; Developmental psychology; Teaching method; Mathematics; Statistics; Vocabulary development","score_opus":0.021624478290005487,"score_gpt":0.3490916083046128,"score_spread":0.3274671300146073,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3113373418","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.998004,0.00009972037,0.0011330363,0.000043667387,0.000006739459,0.000019001207,0.00004417812,0.000008614525,0.00064104004],"genre_scores_gemma":[0.99855703,0.000028881153,0.0011625777,0.00001769986,0.000004284322,0.000030015126,0.000066318,0.000005332454,0.00012778705],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98400474,0.00960501,0.0017257576,0.0010583416,0.003040787,0.00056528323],"domain_scores_gemma":[0.878773,0.101619005,0.0061443197,0.0056866035,0.006244899,0.0015320926],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.023569865,0.0007028677,0.00061901717,0.0028332234,0.0007002875,0.0012905833,0.0006114921,0.00042115897,0.0014349205],"category_scores_gemma":[0.075372994,0.0002847291,0.0015812613,0.001285158,0.0015228862,0.0010697071,0.0017421701,0.0008416782,0.00028175456],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00043397906,0.00007290818,0.9800463,0.00002476775,0.00023401412,0.000052389343,0.0011246168,0.00019302733,0.0016316158,0.00011913012,0.00008754099,0.015979791],"study_design_scores_gemma":[0.000018681154,0.0005629243,0.9944436,0.000021503005,0.00008553814,0.0001871988,0.0008149584,0.0015573568,0.0017893291,0.0001957713,0.00030321107,0.00001985733],"about_ca_topic_score_codex":0.0022793382,"about_ca_topic_score_gemma":0.0032942158,"teacher_disagreement_score":0.023569865,"about_ca_system_score_codex":0.0006343159,"about_ca_system_score_gemma":0.0005814756,"threshold_uncertainty_score":0.124650836},"labels":[],"label_agreement":null}]}