Upload missingness review strict-pairwise assets

.gitattributes

@@ -25,3 +25,4 @@ evaluation/tail/breakdown/v2/prefix_summary__v2.csv filter=lfs diff=lfs merge=lf
 evaluation/query_family/subgroup/data/subgroup_query_rows.csv filter=lfs diff=lfs merge=lfs -text
 evaluation/query_family/subgroup/support3_analysis/query_support3_rows.csv filter=lfs diff=lfs merge=lfs -text
 evaluation/query_family/conditional/breakdown/data/conditional_query_rows.csv filter=lfs diff=lfs merge=lfs -text
+evaluation/query_family/missingness/review_strict_pairwise/data/strict_pairwise_pair_scores.csv filter=lfs diff=lfs merge=lfs -text

evaluation/query_family/missingness/review_strict_pairwise/data/current_vs_strict_pairwise_asset_review.csv

@@ -0,0 +1,87 @@
+dataset_id,model_id,model_label,current_broad_comissing_score,current_marginal_score,current_family_score,strict_status,strict_pairwise_comissing_score,strict_pair_count,active_missing_target_count,delta_strict_minus_current
+c15,arf,ARF,0.994247,0.992016,0.993132,ok,0.991968,253,23,-0.002279
+c15,bayesnet,BayesNet,0.978961,0.969949,0.974455,ok,0.969671,253,23,-0.00929
+c15,realtabformer,RealTabFormer,0.972535,0.960764,0.96665,ok,0.931322,253,23,-0.041213
+c15,tvae,TVAE,0.692437,0.592855,0.642646,ok,0.585239,253,23,-0.107198
+c16,arf,ARF,0.981995,0.990613,0.986304,ok,0.979199,36,9,-0.002796
+c16,bayesnet,BayesNet,0.810089,0.742628,0.776358,ok,0.710908,36,9,-0.099181
+c16,ctgan,CTGAN,0.951149,0.951152,0.951151,ok,0.940609,36,9,-0.01054
+c16,realtabformer,RealTabFormer,0.996877,0.99853,0.997704,ok,0.998207,36,9,0.00133
+c16,tvae,TVAE,0.874315,0.89747,0.885893,ok,0.865348,36,9,-0.008967
+c17,arf,ARF,0.978997,0.997729,0.988363,ok,0.992469,3,3,0.013472
+c17,bayesnet,BayesNet,0.864396,0.836432,0.850414,ok,0.825746,3,3,-0.03865
+c17,ctgan,CTGAN,0.925917,0.963757,0.944837,ok,0.913933,3,3,-0.011984
+c17,realtabformer,RealTabFormer,0.950092,0.966501,0.958297,ok,0.964382,3,3,0.01429
+c17,tvae,TVAE,0.945448,0.94502,0.945234,ok,0.952706,3,3,0.007258
+c17,tabpfgen,TabPFGen,0.864396,0.836432,0.850414,ok,0.825746,3,3,-0.03865
+c18,arf,ARF,0.969362,0.987571,0.978466,ok,0.970942,15,6,0.00158
+c18,bayesnet,BayesNet,0.980698,0.987874,0.984286,ok,0.984106,15,6,0.003408
+c18,realtabformer,RealTabFormer,0.887476,0.867263,0.877369,ok,0.846907,15,6,-0.040569
+c19,arf,ARF,0.998981,0.999878,0.999429,not_applicable_fewer_than_2_missing_targets,,0,1,
+c19,bayesnet,BayesNet,0.994923,0.999328,0.997126,not_applicable_fewer_than_2_missing_targets,,0,1,
+c19,ctgan,CTGAN,0.959485,0.95183,0.955657,not_applicable_fewer_than_2_missing_targets,,0,1,
+c19,realtabformer,RealTabFormer,0.885474,0.853628,0.869551,not_applicable_fewer_than_2_missing_targets,,0,1,
+c19,tvae,TVAE,0.991963,0.990232,0.991097,not_applicable_fewer_than_2_missing_targets,,0,1,
+c19,tabpfgen,TabPFGen,0.987755,0.985958,0.986857,not_applicable_fewer_than_2_missing_targets,,0,1,
+c21,arf,ARF,0.800376,0.720512,0.760444,real_train_csv_missing_locally,,0,0,
+c21,bayesnet,BayesNet,0.800376,0.720512,0.760444,real_train_csv_missing_locally,,0,0,
+c5,arf,ARF,0.99184,0.997438,0.994639,ok,0.990029,6,4,-0.001811
+c5,bayesnet,BayesNet,0.937576,0.997141,0.967358,ok,0.943234,6,4,0.005658
+c5,ctgan,CTGAN,0.929999,0.956625,0.943312,ok,0.930563,6,4,0.000564
+c5,realtabformer,RealTabFormer,0.994431,0.996955,0.995693,ok,0.997044,6,4,0.002613
+c5,tvae,TVAE,0.972853,0.983437,0.978145,ok,0.980151,6,4,0.007298
+c5,tabddpm,TabDDPM,0.803136,0.810198,0.806667,ok,0.725706,6,4,-0.07743
+c5,tabdiff,TabDiff,0.803136,0.810198,0.806667,ok,0.725706,6,4,-0.07743
+c5,tabpfgen,TabPFGen,0.803136,0.810198,0.806667,ok,0.725706,6,4,-0.07743
+c5,tabsyn,TabSyn,0.803136,0.810198,0.806667,ok,0.725706,6,4,-0.07743
+c5,tabsyn,TabSyn,0.803136,0.810198,0.806667,ok,0.725706,6,4,-0.07743
+c5,tabbyflow,TabbyFlow,0.803136,0.810198,0.806667,ok,0.725706,6,4,-0.07743
+m12,arf,ARF,0.603648,0.459932,0.53179,ok,0.495676,1,2,-0.107972
+m12,bayesnet,BayesNet,0.603648,0.459932,0.53179,ok,0.495676,1,2,-0.107972
+m12,ctgan,CTGAN,0.603648,0.459932,0.53179,ok,0.495676,1,2,-0.107972
+m12,realtabformer,RealTabFormer,0.994127,0.99242,0.993273,ok,0.995982,1,2,0.001855
+m12,tvae,TVAE,0.602336,0.465407,0.533872,ok,0.50924,1,2,-0.093096
+m12,tabsyn,TabSyn,0.603648,0.459932,0.53179,ok,0.495676,1,2,-0.107972
+m12,tabbyflow,TabbyFlow,0.603648,0.459932,0.53179,ok,0.495676,1,2,-0.107972
+m7,arf,ARF,0.969658,0.960127,0.964893,not_applicable_fewer_than_2_missing_targets,,0,1,
+m7,bayesnet,BayesNet,0.969658,0.960127,0.964893,not_applicable_fewer_than_2_missing_targets,,0,1,
+m7,ctgan,CTGAN,0.969658,0.960127,0.964893,not_applicable_fewer_than_2_missing_targets,,0,1,
+m7,forestdiffusion,ForestDiffusion,0.969658,0.960127,0.964893,not_applicable_fewer_than_2_missing_targets,,0,1,
+m7,realtabformer,RealTabFormer,0.99502,0.994863,0.994942,not_applicable_fewer_than_2_missing_targets,,0,1,
+m7,tvae,TVAE,0.686329,0.593689,0.640009,not_applicable_fewer_than_2_missing_targets,,0,1,
+m7,tabddpm,TabDDPM,0.969658,0.960127,0.964893,not_applicable_fewer_than_2_missing_targets,,0,1,
+m7,tabdiff,TabDiff,0.969658,0.960127,0.964893,not_applicable_fewer_than_2_missing_targets,,0,1,
+m7,tabpfgen,TabPFGen,0.969658,0.960127,0.964893,not_applicable_fewer_than_2_missing_targets,,0,1,
+m7,tabsyn,TabSyn,0.969658,0.960127,0.964893,not_applicable_fewer_than_2_missing_targets,,0,1,
+m7,tabbyflow,TabbyFlow,0.969658,0.960127,0.964893,not_applicable_fewer_than_2_missing_targets,,0,1,
+m9,arf,ARF,0.986435,0.998453,0.992444,ok,0.987169,21,7,0.000734
+m9,bayesnet,BayesNet,0.99375,0.994743,0.994247,ok,0.995176,21,7,0.001426
+m9,ctgan,CTGAN,0.967434,0.970834,0.969134,ok,0.971807,21,7,0.004373
+m9,realtabformer,RealTabFormer,0.993953,0.994025,0.993989,ok,0.994291,21,7,0.000338
+m9,tvae,TVAE,0.95522,0.971114,0.963167,ok,0.969888,21,7,0.014668
+m9,tabddpm,TabDDPM,0.875667,0.844494,0.860081,ok,0.834484,21,7,-0.041183
+m9,tabpfgen,TabPFGen,0.875667,0.844494,0.860081,ok,0.834484,21,7,-0.041183
+m9,tabsyn,TabSyn,0.875667,0.844494,0.860081,ok,0.834484,21,7,-0.041183
+n13,arf,ARF,0.999738,0.999981,0.999859,real_train_csv_missing_locally,,0,0,
+n13,bayesnet,BayesNet,0.306292,0.009236,0.157764,real_train_csv_missing_locally,,0,0,
+n13,tvae,TVAE,0.306293,0.009236,0.157764,real_train_csv_missing_locally,,0,0,
+n18,arf,ARF,0.854504,0.915683,0.885093,ok,0.841825,14028,168,-0.012679
+n18,bayesnet,BayesNet,0.854504,0.915683,0.885093,ok,0.841825,14028,168,-0.012679
+n18,ctgan,CTGAN,0.854363,0.917687,0.886025,ok,0.843341,14028,168,-0.011022
+n18,tvae,TVAE,0.865029,0.93405,0.89954,ok,0.853099,14028,168,-0.01193
+n4,arf,ARF,0.362426,0.181374,0.2719,ok,0.379146,276,24,0.01672
+n4,bayesnet,BayesNet,0.362426,0.181374,0.2719,ok,0.379146,276,24,0.01672
+n4,ctgan,CTGAN,0.362426,0.181374,0.2719,ok,0.379146,276,24,0.01672
+n4,realtabformer,RealTabFormer,0.977819,0.997466,0.987642,ok,0.993735,276,24,0.015916
+n4,tvae,TVAE,0.482464,0.233281,0.357873,ok,0.662924,276,24,0.18046
+n4,tabddpm,TabDDPM,0.362426,0.181374,0.2719,ok,0.379146,276,24,0.01672
+n4,tabpfgen,TabPFGen,0.362426,0.181374,0.2719,ok,0.379146,276,24,0.01672
+n4,tabsyn,TabSyn,0.362426,0.181374,0.2719,ok,0.379146,276,24,0.01672
+n4,tabbyflow,TabbyFlow,0.362426,0.181374,0.2719,ok,0.379146,276,24,0.01672
+n8,arf,ARF,0.880823,0.855863,0.868343,ok,0.805524,15931,179,-0.075299
+n8,bayesnet,BayesNet,0.880823,0.855863,0.868343,ok,0.805524,15931,179,-0.075299
+n8,ctgan,CTGAN,0.871649,0.845444,0.858546,ok,0.777777,15931,179,-0.093872
+n8,realtabformer,RealTabFormer,0.97491,0.988488,0.981699,ok,0.959667,15931,179,-0.015243
+n8,tvae,TVAE,0.762994,0.692782,0.727888,ok,0.643471,15931,179,-0.119523
+n8,tabsyn,TabSyn,0.880823,0.855863,0.868343,ok,0.805524,15931,179,-0.075299
+n8,tabbyflow,TabbyFlow,0.880823,0.855863,0.868343,ok,0.805524,15931,179,-0.075299

evaluation/query_family/missingness/review_strict_pairwise/data/current_vs_strict_pairwise_model_summary.csv

@@ -0,0 +1,11 @@
+model_id,model_label,dataset_count_overlap,panel_count_overlap,current_broad_comissing_score__mean,strict_pairwise_comissing_score__mean,delta_strict_minus_current__mean
+arf,ARF,10,10,0.860428,0.843395,-0.017033
+bayesnet,BayesNet,10,10,0.826687,0.795101,-0.031586
+ctgan,CTGAN,8,8,0.808323,0.781606,-0.026717
+realtabformer,RealTabFormer,9,9,0.971358,0.964615,-0.006743
+tabbyflow,TabbyFlow,4,4,0.662508,0.601513,-0.060995
+tabddpm,TabDDPM,3,3,0.68041,0.646445,-0.033964
+tabdiff,TabDiff,1,1,0.803136,0.725706,-0.07743
+tabpfgen,TabPFGen,4,4,0.726406,0.691271,-0.035136
+tabsyn,TabSyn,5,6,0.721473,0.66104,-0.060432
+tvae,TVAE,9,9,0.794788,0.78023,-0.014559

evaluation/query_family/missingness/review_strict_pairwise/data/strict_pairwise_dataset_coverage.csv

@@ -0,0 +1,15 @@
+dataset_id,model_panel_count,strict_applicable_panel_count,active_missing_target_count,strict_pair_count
+c15,4,4,23,253
+c16,5,5,9,36
+c17,6,6,3,3
+c18,3,3,6,15
+c19,6,0,1,0
+c21,2,0,0,0
+c5,11,11,4,6
+m12,7,7,2,1
+m7,11,0,1,0
+m9,8,8,7,21
+n13,3,0,0,0
+n18,4,4,168,14028
+n4,9,9,24,276
+n8,7,7,179,15931

evaluation/query_family/missingness/review_strict_pairwise/data/strict_pairwise_pair_scores.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b95db9d1ccc6ef11ebfb41bffabbbdd3bfb9bf5cd6154927d551afecd7a094b3
+size 21047221

evaluation/query_family/missingness/review_strict_pairwise/figures/current_vs_strict_pairwise_review.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c9eacb12234d4861ad3fdf56d56ce9a07a1d966417e57e22c45f4d72c7674df
+size 22797

evaluation/query_family/missingness/review_strict_pairwise/notes/review.md

@@ -0,0 +1,22 @@
+# Strict Pairwise Review
+
+This is a temporary audit only. It does not change the official missingness bundle.
+
+## Audit definition
+
+- Current broad score: official `co_missingness_pattern_consistency` from the direct evaluator.
+- Strict pairwise score: only use unordered pairs of active missing-target columns.
+- For each pair `(A, B)`, score `A | B_missing_indicator` and `B | A_missing_indicator` with the same 0.7 profile + 0.3 strength formula, then average the two directions.
+- Final strict pairwise score = mean over all unordered missing-target pairs.
+
+## Coverage
+
+- Asset/panel rows reviewed: `86`
+- Overlap rows with strict pairwise defined: `64`
+- Datasets with strict pairwise defined: `10`
+- Models with strict pairwise defined: `10`
+
+## Main caveat
+
+- Strict pairwise is undefined when a dataset has fewer than 2 active missing-target columns.
+- So this review is a support-reduced audit, not a drop-in replacement for the official broad score.