Correct polyhedron geometry and line intersection counterfactual

All question and answer/all_answers_combined.csv

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c5bb1bbfa4add115bdf21837504b58f8aea564722760a986fa855da5ccb03963
-size 126966703
+oid sha256:1d2312082155b9ae1132ac12c39d6e6f5ea085020c89a2038afa2d4a48978925
+size 126965335

All question and answer/all_questions_combined.csv

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fc57ada5f1dae80cb78586aea64afbdbfb5fd4a58362147acb8686d75b91baee
+oid sha256:aa3a54051692cbe13725b15dfd9b0be7b3abd145bbb4772a555c5f7b1eba7956
 size 113410628

Dataset/combined_suite_rebuild_report.json

@@ -15,6 +15,7 @@
       "clock_reading": 15000,
       "combination": 15000,
       "combination3d": 15000,
+      "compass_bearing": 15000,
       "coordinate_geometry": 15000,
       "cube_net": 15000,
       "cube_structure": 15000,
@@ -24,16 +25,20 @@
       "fold_punch": 15000,
       "gauge_reading": 15000,
       "gear_train": 15000,
+      "hex_pathfinding": 15000,
       "impossible_object": 15000,
+      "laser_mirror": 15000,
       "line_intersect": 15000,
       "nested_hexagons": 15000,
       "nested_squares": 15000,
       "nested_triangles": 15000,
       "occluded_pattern": 15000,
+      "optical_illusion": 15000,
       "orthographic": 15000,
       "overlap_circles": 15000,
       "physical_stability": 15000,
       "polyhedron": 15000,
+      "projectile_motion": 5000,
       "rotation_match": 15000,
       "route_puzzle": 15000,
       "rpm": 15000,
@@ -41,15 +46,9 @@
       "surface_topology": 15000,
       "symmetry_pattern": 15000
     },
-    "answer_rows": 435000,
+    "answer_rows": 500000,
     "extra_prefixes": [],
-    "missing_prefixes": [
-      "compass_bearing",
-      "hex_pathfinding",
-      "laser_mirror",
-      "optical_illusion",
-      "projectile_motion"
-    ],
+    "missing_prefixes": [],
     "present": true,
     "question_answer_row_counts_match": true,
     "question_prefix_counts": {
@@ -57,6 +56,7 @@
       "clock_reading": 15000,
       "combination": 15000,
       "combination3d": 15000,
+      "compass_bearing": 15000,
       "coordinate_geometry": 15000,
       "cube_net": 15000,
       "cube_structure": 15000,
@@ -66,16 +66,20 @@
       "fold_punch": 15000,
       "gauge_reading": 15000,
       "gear_train": 15000,
+      "hex_pathfinding": 15000,
       "impossible_object": 15000,
+      "laser_mirror": 15000,
       "line_intersect": 15000,
       "nested_hexagons": 15000,
       "nested_squares": 15000,
       "nested_triangles": 15000,
       "occluded_pattern": 15000,
+      "optical_illusion": 15000,
       "orthographic": 15000,
       "overlap_circles": 15000,
       "physical_stability": 15000,
       "polyhedron": 15000,
+      "projectile_motion": 5000,
       "rotation_match": 15000,
       "route_puzzle": 15000,
       "rpm": 15000,
@@ -83,37 +87,10 @@
       "surface_topology": 15000,
       "symmetry_pattern": 15000
     },
-    "question_rows": 435000,
+    "question_rows": 500000,
     "stale_or_mismatched": {
-      "angle_estimation": "question rows differ from current question_set.csv",
-      "clock_reading": "dataset_version is [''], expected legacy-current",
-      "combination": "dataset_version is [''], expected legacy-current",
-      "combination3d": "question rows differ from current question_set.csv",
-      "coordinate_geometry": "dataset_version is [''], expected coordinate-geometry-2.0.0",
-      "cube_net": "question rows differ from current question_set.csv",
-      "cube_structure": "question rows differ from current question_set.csv",
-      "depth_height": "dataset_version is [''], expected depth-height-2.0.0",
-      "embedded_figures": "dataset_version is [''], expected legacy-current",
-      "fbd": "question rows differ from current question_set.csv",
-      "fold_punch": "question rows differ from current question_set.csv",
-      "gauge_reading": "dataset_version is [''], expected legacy-current",
-      "gear_train": "dataset_version is [''], expected legacy-current",
-      "impossible_object": "question rows differ from current question_set.csv",
-      "line_intersection": "dataset_version is [''], expected line-intersection-2.0.0",
-      "nested_hexagons": "question rows differ from current question_set.csv",
-      "nested_squares": "question rows differ from current question_set.csv",
-      "nested_triangles": "question rows differ from current question_set.csv",
-      "occluded_pattern": "dataset_version is [''], expected occluded-pattern-2.0.0",
-      "orthographic": "dataset_version is [''], expected legacy-current",
-      "overlap_circles": "question rows differ from current question_set.csv",
-      "physical_stability": "dataset_version is [''], expected legacy-current",
-      "polyhedron": "dataset_version is [''], expected polyhedron-3.0.0",
-      "rotation_matching": "question rows differ from current question_set.csv",
-      "route": "dataset_version is [''], expected route-2.0.0",
-      "rpm": "question rows differ from current question_set.csv",
-      "shadow_inference": "question rows differ from current question_set.csv",
-      "surface_topology": "question rows differ from current question_set.csv",
-      "symmetry_pattern": "dataset_version is [''], expected symmetry-pattern-2.0.0"
+      "line_intersection": "question rows differ from current question_set.csv",
+      "polyhedron": "dataset_version is ['polyhedron-3.0.0'], expected polyhedron-4.0.0"
     }
   },
   "discovered_dataset_count": 34,
@@ -266,7 +243,7 @@
       "prefix": "line_intersect",
       "questions": 15000,
       "slug": "line_intersection",
-      "version": "line-intersection-2.0.0"
+      "version": "line-intersection-3.0.0"
     },
     "nested_hexagons_dataset_3000": {
       "answers": 15000,
@@ -338,7 +315,7 @@
       "prefix": "polyhedron",
       "questions": 15000,
       "slug": "polyhedron",
-      "version": "polyhedron-3.0.0"
+      "version": "polyhedron-4.0.0"
     },
     "projectile_motion_dataset_1000": {
       "answers": 5000,

Dataset/final_suite_audit.json

@@ -18660,7 +18660,7 @@
         "p95": 4.0
       }
     },
-    "dataset_version": "line-intersection-2.0.0",
+    "dataset_version": "line-intersection-3.0.0",
     "degeneracy_flags": {},
     "features_at_v_ge_0_10": {
       "crossed_from_above_to_below": {
@@ -18668,53 +18668,55 @@
         "2": 0.99933267,
         "3": 0.24382486,
         "4": 0.9974952,
-        "5": 0.65641789
+        "5": 0.10633523
       },
       "difficulty_score": {
         "1": 0.85249175,
         "2": 0.70761258,
         "3": 0.18609249,
         "4": 0.48280098,
-        "5": 0.59575224
+        "5": 0.27643382
       },
       "leftmost_intersection_x": {
         "1": 0.17244787,
         "2": 0.50316301,
         "3": 0.21808086,
         "4": 0.46245306,
-        "5": 0.50316301
+        "5": 0.20153092
       },
       "num_blue_segments": {
         "1": 1.0,
         "2": 0.35195787,
         "4": 0.34851292,
-        "5": 0.3207622
+        "5": 0.24488677
       },
       "num_red_segments": {
         "1": 1.0,
         "2": 0.35195787,
         "4": 0.34851292,
-        "5": 0.3207622
+        "5": 0.24488677
       },
       "red_above_blue_at_end": {
-        "3": 0.62772958
+        "3": 0.62772958,
+        "5": 0.19278923
       },
       "red_above_blue_at_start": {
-        "3": 0.63335914
+        "3": 0.63335914,
+        "5": 0.19749312
       },
       "rightmost_intersection_x": {
         "1": 0.17147737,
         "2": 0.502831,
         "3": 0.21814305,
         "4": 0.46222129,
-        "5": 0.502831
+        "5": 0.20119416
       },
       "total_intersections": {
         "1": 0.32378613,
         "2": 0.99983304,
         "3": 0.23746468,
         "4": 0.86218295,
-        "5": 0.82743926
+        "5": 0.31018458
       }
     },
     "images": 3000,
@@ -18765,7 +18767,7 @@
             "kind": "categorical"
           },
           "5": {
-            "cramers_v": 0.6564178921365604,
+            "cramers_v": 0.10633522642820674,
             "kind": "categorical"
           }
         },
@@ -18791,7 +18793,7 @@
             "kind": "numeric_deciles"
           },
           "5": {
-            "cramers_v": 0.5957522424527353,
+            "cramers_v": 0.2764338162488935,
             "kind": "numeric_deciles"
           }
         },
@@ -18817,7 +18819,7 @@
             "kind": "categorical"
           },
           "5": {
-            "cramers_v": 0.5031630080720518,
+            "cramers_v": 0.20153091676455742,
             "kind": "categorical"
           }
         },
@@ -18843,7 +18845,7 @@
             "kind": "numeric_deciles"
           },
           "5": {
-            "cramers_v": 0.32076220310426695,
+            "cramers_v": 0.2448867721693077,
             "kind": "numeric_deciles"
           }
         },
@@ -18869,7 +18871,7 @@
             "kind": "numeric_deciles"
           },
           "5": {
-            "cramers_v": 0.32076220310426695,
+            "cramers_v": 0.2448867721693077,
             "kind": "numeric_deciles"
           }
         },
@@ -18895,7 +18897,7 @@
             "kind": "categorical"
           },
           "5": {
-            "cramers_v": 0.024939328296776968,
+            "cramers_v": 0.19278923326330089,
             "kind": "categorical"
           }
         },
@@ -18921,7 +18923,7 @@
             "kind": "categorical"
           },
           "5": {
-            "cramers_v": 0.027270754941034103,
+            "cramers_v": 0.19749312158764903,
             "kind": "categorical"
           }
         },
@@ -18973,7 +18975,7 @@
             "kind": "categorical"
           },
           "5": {
-            "cramers_v": 0.5028309968131334,
+            "cramers_v": 0.2011941575669298,
             "kind": "categorical"
           }
         },
@@ -18999,7 +19001,7 @@
             "kind": "numeric_deciles"
           },
           "5": {
-            "cramers_v": 0.8274392623607943,
+            "cramers_v": 0.3101845784550317,
             "kind": "numeric_deciles"
           }
         },
@@ -19087,14 +19089,14 @@
         }
       },
       "5": {
-        "constant_answer_baseline": 0.2976666666666667,
+        "constant_answer_baseline": 0.49633333333333335,
         "counts": {
-          "0": 858,
-          "1": 893,
-          "2": 683,
-          "3": 332,
-          "4": 167,
-          "5": 67
+          "0": 1489,
+          "1": 764,
+          "2": 557,
+          "3": 130,
+          "4": 58,
+          "5": 2
         }
       }
     },
@@ -26324,34 +26326,31 @@
     "categorical_distributions": {
       "face_shape_types": {
         "mixed": 1738,
-        "pentagons": 472,
+        "pentagons": 315,
         "squares": 158,
-        "triangles": 632
+        "triangles": 789
       },
       "is_convex": {
-        "False": 630,
-        "True": 2370
+        "False": 316,
+        "True": 2684
       },
       "solid_class": {
         "Archimedean": 948,
         "Catalan": 632,
         "Compound": 316,
-        "NonConvex": 314,
-        "Platonic": 790
+        "Platonic": 1104
       },
       "solid_name": {
         "compound of cube and octahedron": 158,
         "cube": 158,
         "cuboctahedron": 158,
-        "dodecahedron": 158,
-        "great dodecahedron": 157,
-        "icosahedron": 158,
+        "dodecahedron": 315,
+        "icosahedron": 315,
         "icosidodecahedron": 158,
         "octahedron": 158,
         "rhombic dodecahedron": 158,
         "rhombic triacontahedron": 158,
         "rhombicuboctahedron": 158,
-        "small stellated dodecahedron": 157,
         "stella octangula": 158,
         "tetrahedron": 158,
         "triakis octahedron": 158,
@@ -26376,7 +26375,7 @@
         "max": 32.0,
         "min": 4.0,
         "p25": 8.0,
-        "p50": 12.0,
+        "p50": 14.0,
         "p75": 20.0,
         "p95": 32.0
       },
@@ -26399,13 +26398,13 @@
         "p95": 16.0
       }
     },
-    "dataset_version": "polyhedron-3.0.0",
+    "dataset_version": "polyhedron-4.0.0",
     "degeneracy_flags": {
       "2": {
-        "constant_answer_baseline": 0.79,
+        "constant_answer_baseline": 0.8946666666666667,
         "counts": {
-          "convex": 2370,
-          "non-convex": 630
+          "convex": 2684,
+          "non-convex": 316
         }
       },
       "5": {
@@ -26417,52 +26416,52 @@
     },
     "features_at_v_ge_0_10": {
       "edge_count": {
-        "1": 0.75583117,
-        "2": 0.48189444,
-        "3": 0.67088382,
-        "4": 0.5883061
+        "1": 0.76787728,
+        "2": 0.45461294,
+        "3": 0.65853928,
+        "4": 0.57636344
       },
       "face_count": {
-        "1": 0.9994987,
-        "2": 0.41087482,
-        "3": 0.70705014,
-        "4": 0.70138458
+        "1": 0.99966572,
+        "2": 0.45387864,
+        "3": 0.70242336,
+        "4": 0.67866732
       },
       "face_shape_types": {
-        "1": 0.86719078,
-        "2": 0.51107672,
-        "3": 0.8201062,
-        "4": 0.65396173
+        "1": 0.85781495,
+        "2": 0.20976386,
+        "3": 0.85076687,
+        "4": 0.58908833
       },
       "is_convex": {
-        "1": 0.40965529,
+        "1": 0.45314314,
         "2": 1.0,
-        "3": 0.42016178,
-        "4": 0.82962824
+        "3": 0.36265877,
+        "4": 0.99766238
       },
       "solid_class": {
-        "1": 0.68039958,
-        "2": 0.99949954,
-        "3": 0.61202071,
-        "4": 0.82161564
+        "1": 0.72115872,
+        "2": 0.99966639,
+        "3": 0.63374451,
+        "4": 0.86455964
       },
       "solid_name": {
-        "1": 0.99849385,
-        "2": 0.99716075,
-        "3": 0.55664183,
-        "4": 0.61491684
+        "1": 0.99882875,
+        "2": 0.9974952,
+        "3": 0.59153212,
+        "4": 0.60079841
       },
       "vertex_count": {
-        "1": 0.77114061,
-        "2": 0.42338604,
-        "3": 0.75755222,
-        "4": 0.61836357
+        "1": 0.80136852,
+        "2": 0.50258694,
+        "3": 0.76737935,
+        "4": 0.61153135
       },
       "visible_face_count": {
-        "1": 0.94450613,
-        "2": 0.38803547,
-        "3": 0.70008253,
-        "4": 0.67368945
+        "1": 0.98613929,
+        "2": 0.44154719,
+        "3": 0.70777868,
+        "4": 0.66677894
       }
     },
     "images": 3000,
@@ -26471,19 +26470,19 @@
         "classification": "scene_feature",
         "levels": {
           "1": {
-            "cramers_v": 0.7558311692393166,
+            "cramers_v": 0.7678772775143334,
             "kind": "numeric_deciles"
           },
           "2": {
-            "cramers_v": 0.48189444296185086,
+            "cramers_v": 0.45461294361222804,
             "kind": "numeric_deciles"
           },
           "3": {
-            "cramers_v": 0.6708838217815941,
+            "cramers_v": 0.6585392806416459,
             "kind": "numeric_deciles"
           },
           "4": {
-            "cramers_v": 0.5883060976992267,
+            "cramers_v": 0.5763634416760484,
             "kind": "numeric_deciles"
           },
           "5": {
@@ -26497,19 +26496,19 @@
         "classification": "scene_feature",
         "levels": {
           "1": {
-            "cramers_v": 0.999498704956384,
+            "cramers_v": 0.999665719529353,
             "kind": "numeric_deciles"
           },
           "2": {
-            "cramers_v": 0.4108748230510013,
+            "cramers_v": 0.4538786369638581,
             "kind": "numeric_deciles"
           },
           "3": {
-            "cramers_v": 0.7070501355062565,
+            "cramers_v": 0.7024233562734422,
             "kind": "numeric_deciles"
           },
           "4": {
-            "cramers_v": 0.701384580685305,
+            "cramers_v": 0.6786673193089476,
             "kind": "numeric_deciles"
           },
           "5": {
@@ -26523,19 +26522,19 @@
         "classification": "scene_feature",
         "levels": {
           "1": {
-            "cramers_v": 0.867190777357142,
+            "cramers_v": 0.8578149524363928,
             "kind": "categorical"
           },
           "2": {
-            "cramers_v": 0.5110767202699095,
+            "cramers_v": 0.20976385557014507,
             "kind": "categorical"
           },
           "3": {
-            "cramers_v": 0.8201061981936764,
+            "cramers_v": 0.8507668742491247,
             "kind": "categorical"
           },
           "4": {
-            "cramers_v": 0.6539617344974008,
+            "cramers_v": 0.5890883344193385,
             "kind": "categorical"
           },
           "5": {
@@ -26549,19 +26548,19 @@
         "classification": "scene_feature",
         "levels": {
           "1": {
-            "cramers_v": 0.40965528570468135,
+            "cramers_v": 0.4531431403922659,
             "kind": "categorical"
           },
           "2": {
-            "cramers_v": 0.9999999999999999,
+            "cramers_v": 1.0,
             "kind": "categorical"
           },
           "3": {
-            "cramers_v": 0.4201617750145626,
+            "cramers_v": 0.3626587678392811,
             "kind": "categorical"
           },
           "4": {
-            "cramers_v": 0.8296282369790656,
+            "cramers_v": 0.9976623778346884,
             "kind": "categorical"
           },
           "5": {
@@ -26575,19 +26574,19 @@
         "classification": "scene_feature",
         "levels": {
           "1": {
-            "cramers_v": 0.6803995831650559,
+            "cramers_v": 0.7211587248153378,
             "kind": "categorical"
           },
           "2": {
-            "cramers_v": 0.9994995412147983,
+            "cramers_v": 0.9996663886479304,
             "kind": "categorical"
           },
           "3": {
-            "cramers_v": 0.6120207081547728,
+            "cramers_v": 0.6337445100277069,
             "kind": "categorical"
           },
           "4": {
-            "cramers_v": 0.821615636025442,
+            "cramers_v": 0.8645596407847268,
             "kind": "categorical"
           },
           "5": {
@@ -26601,19 +26600,19 @@
         "classification": "scene_feature",
         "levels": {
           "1": {
-            "cramers_v": 0.9984938490322233,
+            "cramers_v": 0.9988287455195981,
             "kind": "categorical"
           },
           "2": {
-            "cramers_v": 0.9971607458355733,
+            "cramers_v": 0.9974951951979328,
             "kind": "categorical"
           },
           "3": {
-            "cramers_v": 0.5566418257069514,
+            "cramers_v": 0.5915321198219095,
             "kind": "categorical"
           },
           "4": {
-            "cramers_v": 0.6149168351347316,
+            "cramers_v": 0.6007984103612805,
             "kind": "categorical"
           },
           "5": {
@@ -26627,19 +26626,19 @@
         "classification": "scene_feature",
         "levels": {
           "1": {
-            "cramers_v": 0.7711406057961122,
+            "cramers_v": 0.8013685193945911,
             "kind": "numeric_deciles"
           },
           "2": {
-            "cramers_v": 0.4233860362173674,
+            "cramers_v": 0.502586939464905,
             "kind": "numeric_deciles"
           },
           "3": {
-            "cramers_v": 0.7575522205488738,
+            "cramers_v": 0.7673793499859815,
             "kind": "numeric_deciles"
           },
           "4": {
-            "cramers_v": 0.6183635665536725,
+            "cramers_v": 0.6115313451725574,
             "kind": "numeric_deciles"
           },
           "5": {
@@ -26653,19 +26652,19 @@
         "classification": "scene_feature",
         "levels": {
           "1": {
-            "cramers_v": 0.9445061270013774,
+            "cramers_v": 0.9861392881588266,
             "kind": "numeric_deciles"
           },
           "2": {
-            "cramers_v": 0.3880354657584141,
+            "cramers_v": 0.4415471852371717,
             "kind": "numeric_deciles"
           },
           "3": {
-            "cramers_v": 0.7000825297107476,
+            "cramers_v": 0.7077786799672889,
             "kind": "numeric_deciles"
           },
           "4": {
-            "cramers_v": 0.6736894549459747,
+            "cramers_v": 0.6667789386007642,
             "kind": "numeric_deciles"
           },
           "5": {
@@ -26678,11 +26677,11 @@
     },
     "level_distributions": {
       "1": {
-        "constant_answer_baseline": 0.26266666666666666,
+        "constant_answer_baseline": 0.21066666666666667,
         "counts": {
-          "12": 788,
+          "12": 631,
           "14": 632,
-          "20": 158,
+          "20": 315,
           "24": 158,
           "26": 158,
           "30": 158,
@@ -26693,10 +26692,10 @@
         }
       },
       "2": {
-        "constant_answer_baseline": 0.79,
+        "constant_answer_baseline": 0.8946666666666667,
         "counts": {
-          "convex": 2370,
-          "non-convex": 630
+          "convex": 2684,
+          "non-convex": 316
         }
       },
       "3": {
@@ -26721,9 +26720,9 @@
           "7": 205,
           "8": 158,
           "mixed": 585,
-          "pentagons": 178,
+          "pentagons": 120,
           "squares": 51,
-          "triangles": 205
+          "triangles": 263
         }
       },
       "4": {
@@ -26742,8 +26741,7 @@
           "Archimedean": 273,
           "Catalan": 206,
           "Compound": 97,
-          "NonConvex": 112,
-          "Platonic": 207,
+          "Platonic": 319,
           "no": 826,
           "yes": 219
         }

Dataset/final_suite_audit.md

@@ -19,7 +19,7 @@
 | hex_pathfinding_dataset_3000 | hex-pathfinding-2.0.0 | yes | 0 | 11.1% | 50.0% | 21.6% | 22.0% | 33.3% | 3000/3000 |
 | impossible_object_dataset_3000 | impossible-object-4.0.0 | no | 0 | 50.1% | 50.0% | 20.7% | 25.0% | 50.0% | 3000/3000 |
 | laser_mirror_dataset_3000 | legacy-current | no | 0 | 25.1% | 75.0% | 52.3% | 4.4% | 3.7% | 3000/3000 |
-| line_intersection_dataset_3000 | line-intersection-2.0.0 | yes | 0 | 40.8% | 27.3% | 20.8% | 19.3% | 29.8% | 3000/3000 |
+| line_intersection_dataset_3000 | line-intersection-3.0.0 | yes | 0 | 40.8% | 27.3% | 20.8% | 19.3% | 49.6% | 3000/3000 |
 | nested_hexagons_dataset_3000 | nested-hexagons-8.0.0 | no | 0 | 11.5% | 50.0% | 5.2% | 15.0% | 50.0% | 3000/3000 |
 | nested_squares_dataset_3000 | nested-squares-8.0.0 | no | 0 | 11.5% | 50.0% | 5.2% | 15.0% | 50.0% | 3000/3000 |
 | nested_triangles_dataset_3000 | nested-triangles-8.0.0 | no | 0 | 11.5% | 50.0% | 5.1% | 15.0% | 50.0% | 3000/3000 |
@@ -28,7 +28,7 @@
 | orthographic_dataset_3000 | legacy-current | no | 0 | 23.5% | 18.8% | 32.8% | 46.5% | 100.0% | 3000/3000 |
 | overlap_circles_dataset_3000 | overlap-circles-2.0.0 | yes | 0 | 13.3% | 36.1% | 34.5% | 21.6% | 9.3% | 3000/3000 |
 | physical_stability_dataset_3000 | legacy-current | no | 0 | 30.8% | 62.5% | 30.5% | 50.0% | 16.6% | 3000/3000 |
-| polyhedron_dataset_3000 | polyhedron-3.0.0 | yes | 0 | 26.3% | 79.0% | 19.5% | 27.5% | 100.0% | 3000/3000 |
+| polyhedron_dataset_3000 | polyhedron-4.0.0 | yes | 0 | 21.1% | 89.5% | 19.5% | 27.5% | 100.0% | 3000/3000 |
 | projectile_motion_dataset_1000 | projectile-motion-1.0.0 | no | 0 | 2.8% | 65.4% | 1.0% | 0.4% | 68.8% | 1000/1000 |
 | rotation_matching_dataset_3000 | rotation-matching-2.0.0 | no | 0 | 25.8% | 25.0% | 15.2% | 26.0% | 14.3% | 3000/3000 |
 | route_dataset_3000 | route-2.0.0 | yes | 0 | 23.4% | 51.3% | 11.3% | 3.5% | 26.3% | 3000/3000 |

Dataset/line_intersection_dataset_3000/README.md

@@ -2,7 +2,13 @@
 
 ## 1. Dataset overview
 
-This dataset contains **3,000 images** and **12,000 questions**. It was generated on **2026-08-10** using a fully synthetic, deterministic Python pipeline; no image was human-drawn and no answer was manually labeled. The visual design and dual-wording Level-2 format were adapted from an existing published methodology for testing line-intersection counting in vision-language models. This dataset extends that design with a four-level difficulty structure, variable segment counts, explicit self-intersection handling, and coordinate-derived compound questions.
+The current v3 dataset contains **3,000 images** and **15,000 questions** (five ordered levels per image). It was generated using a fully synthetic, deterministic Python pipeline; no image was human-drawn and no answer was manually labeled. The visual design and dual-wording Level-2 format were adapted from an existing published methodology for testing line-intersection counting in vision-language models.
+
+### Version 3 Level-5 counterfactual correction
+
+The former 20-pixel downward translation was independently recomputed and found to preserve the intersection count in all 2,016 translation-template items because it was smaller than the generator's endpoint separation. Version 3 uses 60 pixels and always rebuilds the translated red coordinates before running the full red-blue segment-intersection search. This changes 1,400 Level-5 answers; the translated-minus-original distribution is recorded in `validation_metrics.json`. The other 984 Level-5 items retain the remove-first-red-segment template. The new overall Level-5 constant-answer baseline is 49.63%.
+
+All five levels, point/segment and intersection-array invariants, leak statistics, guard injections, and all 3,000 PNGs are exhaustively revalidated. The long verbatim prompt below is retained as historical design provenance; the release description above and v3 generator are authoritative.
 
 ## 2. Full verbatim generation prompt
 

Dataset/line_intersection_dataset_3000/build_manifest.json

@@ -1,9 +1,20 @@
 {
-  "constraint_source": "generator constants and validation metrics",
+  "constraint_set": {
+    "blue_points_length_equals_num_blue_segments_plus_one": true,
+    "five_ordered_questions_per_image": true,
+    "intersection_array_length_equals_total_intersections": true,
+    "level5_translation_pixels": 60,
+    "level5_translation_reintersects_from_transformed_coordinates": true,
+    "no_endpoint_or_collinear_degenerate_intersections": true,
+    "png_red_blue_polyline_recovery_all_items": true,
+    "red_points_length_equals_num_red_segments_plus_one": true
+  },
+  "constraint_source": "generator constants and exhaustive validation metrics",
   "current_build_layout": "unsuffixed",
-  "dataset_version": "line-intersection-2.0.0",
-  "generator_commit": "e6975d7f82e6fc42f77fdc6434e733c10adb83d6",
+  "dataset_version": "line-intersection-3.0.0",
+  "generator_commit": "c0bf63192b57ed5f7269adce67557adcebddd323",
   "images": 3000,
+  "metadata_only_rebuild": true,
   "question_set_fields": [
     "question_id",
     "task",

Dataset/line_intersection_dataset_3000/flatten_annotations.py

@@ -2,18 +2,18 @@ from __future__ import annotations
 import argparse,csv,json
 from collections import Counter
 from pathlib import Path
-TASKS={1:"Image Description",2:"Basic Relational Reasoning",3:"Comparative Reasoning",4:"Compound Reasoning"};FIELDS=["difficulty_score","total_intersections","num_red_segments","num_blue_segments","wording_variant","seed"]
+TASKS={1:"Image Description",2:"Basic Relational Reasoning",3:"Comparative Reasoning",4:"Compound Reasoning",5:"Extrapolative/Counterfactual Reasoning"};FIELDS=["difficulty_score","total_intersections","num_red_segments","num_blue_segments","wording_variant","seed"]
 def main():
  p=argparse.ArgumentParser();p.add_argument("annotations",type=Path);a=p.parse_args();rows=[];images=0;bad=[]
  for line in a.annotations.read_text(encoding="utf-8").splitlines():
   if not line:continue
   r=json.loads(line);images+=1
-  if len(r["questions"])!=4:bad.append(r["id"])
+  if len(r["questions"])!=5:bad.append(r["id"])
   meta=json.dumps({k:r[k] for k in FIELDS},sort_keys=True,separators=(",",":"))
   for q in r["questions"]:rows.append({"task":TASKS[q["difficulty_level"]],"image":Path(r["image_path"]).name,"prompt":q["question_text"],"groundtruth":str(q["ground_truth"]),"metadata":meta})
  out=a.annotations.parent;cols=["task","image","prompt","groundtruth","metadata"]
  with (out/"dataset_final.csv").open("w",encoding="utf-8-sig",newline="") as f:w=csv.DictWriter(f,fieldnames=cols,lineterminator="\n");w.writeheader();w.writerows(rows)
  with (out/"dataset_final.jsonl").open("w",encoding="utf-8",newline="\n") as f:
   for r in rows:f.write(json.dumps(r,separators=(",",":"))+"\n")
- counts=Counter(r["task"] for r in rows);print(f"Images: {images}\nRows: {len(rows)}\nExpected: {4*images}\nImages without 4 questions: {len(bad)}");[print(f"{x}: {counts[x]}") for x in TASKS.values()];assert len(rows)==4*images and not bad;print("Sanity check: PASS")
+ counts=Counter(r["task"] for r in rows);print(f"Images: {images}\nRows: {len(rows)}\nExpected: {5*images}\nImages without 5 questions: {len(bad)}");[print(f"{x}: {counts[x]}") for x in TASKS.values()];assert len(rows)==5*images and not bad;print("Sanity check: PASS")
 if __name__=="__main__":main()

Dataset/line_intersection_dataset_3000/generate_line_intersection_dataset.py

@@ -8,6 +8,7 @@ try:
 except ImportError:
  def tqdm(x,**_):return x
 BG="#FDFBF6";RED="#D7192D";BLUE="#1769B0";AA=3;EPS=1e-9
+VERSION="line-intersection-3.0.0";LEVEL5_TRANSLATION_PX=60
 def segment_intersection(p1,p2,p3,p4):
  x1,y1=p1;x2,y2=p2;x3,y3=p3;x4,y4=p4;rx=x2-x1;ry=y2-y1;sx=x4-x3;sy=y4-y3;den=rx*sy-ry*sx
  if abs(den)<EPS:return None
@@ -66,11 +67,17 @@ def questions(iid,meta,rng):
  if t=="parity":text="If the two lines intersect an odd number of times, the line that starts higher must end lower (and vice versa). Based on this rule, do these two lines intersect an odd or even number of times? Answer 'odd' or 'even'.";gt="odd" if meta["crossed_from_above_to_below"] else "even";typ="intersection_parity";fmt="choice"
  elif t=="halves":text="How many intersection points occur in the left half of the image versus the right half? Answer as two numbers separated by a comma, e.g. {2,1}.";left=sum(p["x"]<meta["canvas_size"][0]/2 for p in meta["intersections"]);gt=f"{left},{meta['total_intersections']-left}";typ="intersection_half_counts";fmt="numeric_pair"
  else:text="If you removed the segment of the red line closest to the left edge, how many intersection points would remain? Answer with a number in curly brackets.";gt=str(meta["total_intersections"]-sum(p["red_segment_index"]==0 for p in meta["intersections"]));typ="remove_first_red_segment";fmt="numeric"
- qs.append({"question_id":iid+"_q4","question_text":text,"question_type":typ,"ground_truth":gt,"answer_format":fmt,"difficulty_level":4});return qs
+ qs.append({"question_id":iid+"_q4","question_text":text,"question_type":typ,"ground_truth":gt,"answer_format":fmt,"difficulty_level":4})
+ if rng.random()<2/3:
+  moved=[[x,y+LEVEL5_TRANSLATION_PX] for x,y in meta["red_points"]]
+  text=f"If the entire red polyline were translated exactly {LEVEL5_TRANSLATION_PX} pixels downward without changing its shape, how many red-blue intersections would remain?";gt=str(len(compute_intersections(moved,meta["blue_points"])));typ="translate_red_intersections"
+ else:
+  text="If you removed the segment of the red line closest to the left edge, how many intersection points would remain? Answer with a number in curly brackets.";gt=str(meta["total_intersections"]-sum(p["red_segment_index"]==0 for p in meta["intersections"]));typ="remove_first_red_segment"
+ qs.append({"question_id":iid+"_q5","question_text":text,"question_type":typ,"ground_truth":gt,"answer_format":"numeric","difficulty_level":5});return qs
 def render(path,size,red,blue):
  w,h=size;im=Image.new("RGB",(w*AA,h*AA),BG);d=ImageDraw.Draw(im);d.line([(x*AA,y*AA) for x,y in red],fill=RED,width=6,joint="curve");d.line([(x*AA,y*AA) for x,y in blue],fill=BLUE,width=6,joint="curve");im.resize((w,h),Image.Resampling.LANCZOS).save(path,"PNG")
 def generate_one(i,images):
- rng,size,red,blue,ints=generate_geometry(i);iid=f"line_intersect_{i:04d}";render(images/f"{iid}.png",size,red,blue);total=len(ints);meta={"canvas_size":list(size),"num_red_segments":len(red)-1,"num_blue_segments":len(blue)-1,"total_intersections":total,"intersections":ints,"leftmost_intersection_x":ints[0]["x"] if ints else None,"rightmost_intersection_x":ints[-1]["x"] if ints else None,"red_above_blue_at_start":red[0][1]<blue[0][1],"red_above_blue_at_end":red[-1][1]<blue[-1][1],"crossed_from_above_to_below":(red[0][1]<blue[0][1])!=(red[-1][1]<blue[-1][1]),"red_self_intersecting":is_self_intersecting(red),"blue_self_intersecting":is_self_intersecting(blue),"seed":i,"red_points":red,"blue_points":blue};difficulty=round(min(1,.45*total/5+.35*((len(red)-1)+(len(blue)-1)-4)/6+.2*(total>2)),4);row={"id":iid,"image_path":f"images/{iid}.png",**meta,"difficulty_score":difficulty};row["questions"]=questions(iid,row,rng);return row
+ rng,size,red,blue,ints=generate_geometry(i);iid=f"line_intersect_{i:04d}";render(images/f"{iid}.png",size,red,blue);total=len(ints);meta={"canvas_size":list(size),"num_red_segments":len(red)-1,"num_blue_segments":len(blue)-1,"total_intersections":total,"intersections":ints,"leftmost_intersection_x":ints[0]["x"] if ints else None,"rightmost_intersection_x":ints[-1]["x"] if ints else None,"red_above_blue_at_start":red[0][1]<blue[0][1],"red_above_blue_at_end":red[-1][1]<blue[-1][1],"crossed_from_above_to_below":(red[0][1]<blue[0][1])!=(red[-1][1]<blue[-1][1]),"red_self_intersecting":is_self_intersecting(red),"blue_self_intersecting":is_self_intersecting(blue),"seed":i,"red_points":red,"blue_points":blue,"dataset_version":VERSION,"frame_conventions":{"stored_geometry":"image_pixel_frame","render":"image_pixel_frame","level5_translation":f"red y + {LEVEL5_TRANSLATION_PX} pixels"}};difficulty=round(min(1,.45*total/5+.35*((len(red)-1)+(len(blue)-1)-4)/6+.2*(total>2)),4);row={"id":iid,"image_path":f"images/{iid}.png",**meta,"difficulty_score":difficulty};row["questions"]=questions(iid,row,rng);return row
 def generate_dataset(n=3000,output_dir="line_intersection_dataset"):
  out=Path(output_dir);images=out/"images";images.mkdir(parents=True,exist_ok=True)
  with (out/"annotations.jsonl").open("w",encoding="utf-8",newline="\n") as f:

Dataset/line_intersection_dataset_3000/validate_line_intersect_dataset.py

@@ -1,26 +1,54 @@
+"""Exhaustive five-level and PNG validator for line-intersection v3."""
 from __future__ import annotations
-import argparse,json,math
+import argparse,json,sys
+from collections import Counter
 from pathlib import Path
-from generate_line_intersection_dataset import compute_intersections,is_self_intersecting
+import numpy as np
+from PIL import Image,ImageDraw
+from generate_line_intersection_dataset import BG,BLUE,RED,compute_intersections,is_self_intersecting
+sys.path.insert(0,str(Path(__file__).resolve().parents[1]))
+from benchmark_validation_utils import answer_distributions,leak_audit,quantiles
+VERSION="line-intersection-3.0.0";SHIFT=60;AA=3
+def png_matches(root,row):
+ w,h=row["canvas_size"];ref=Image.new("RGB",(w*AA,h*AA),BG);draw=ImageDraw.Draw(ref);draw.line([(x*AA,y*AA) for x,y in row["red_points"]],fill=RED,width=6,joint="curve");draw.line([(x*AA,y*AA) for x,y in row["blue_points"]],fill=BLUE,width=6,joint="curve");ref=ref.resize((w,h),Image.Resampling.LANCZOS)
+ with Image.open(root/row["image_path"]) as actual:return list(actual.size)==row["canvas_size"] and np.array_equal(np.asarray(actual.convert("RGB")),np.asarray(ref))
+def derived_answers(row,intersections):
+ red,blue=row["red_points"],row["blue_points"];left=sum(p["x"]<row["canvas_size"][0]/2 for p in intersections);moved=[[x,y+SHIFT] for x,y in red]
+ return {"red_segment_count":str(len(red)-1),"blue_segment_count":str(len(blue)-1),"intersection_count":str(len(intersections)),"higher_at_start":"red" if red[0][1]<blue[0][1] else "blue","red_end_relation":"above" if red[-1][1]<blue[-1][1] else "below","leftmost_intersection_side":("left" if intersections[0]["x"]<row["canvas_size"][0]/2 else "right") if intersections else None,"intersection_parity":"odd" if (red[0][1]<blue[0][1])!=(red[-1][1]<blue[-1][1]) else "even","intersection_half_counts":f"{left},{len(intersections)-left}","remove_first_red_segment":str(len(intersections)-sum(p["red_segment_index"]==0 for p in intersections)),"translate_red_intersections":str(len(compute_intersections(moved,blue)))}
 def validate(root):
- issues=[];checked=0
- for line in (root/"annotations.jsonl").read_text(encoding="utf-8").splitlines():
-  if not line:continue
-  r=json.loads(line);checked+=1;iid=r["id"];red=r["red_points"];blue=r["blue_points"]
-  if not (root/r["image_path"]).is_file():issues.append(f"{iid}: missing image")
-  try:ints=compute_intersections(red,blue,True)
-  except ValueError as e:issues.append(f"{iid}: degenerate intersection: {e}");continue
-  if len(ints)!=r["total_intersections"]:issues.append(f"{iid}: intersection count mismatch")
-  if ints!=r["intersections"]:issues.append(f"{iid}: intersection coordinates mismatch")
-  derived={"red_segment_count":str(len(red)-1),"blue_segment_count":str(len(blue)-1),"intersection_count":str(len(ints)),"higher_at_start":"red" if red[0][1]<blue[0][1] else "blue","red_end_relation":"above" if red[-1][1]<blue[-1][1] else "below","intersection_parity":"odd" if (red[0][1]<blue[0][1])!=(red[-1][1]<blue[-1][1]) else "even","remove_first_red_segment":str(len(ints)-sum(p["red_segment_index"]==0 for p in ints))}
-  derived["leftmost_intersection_side"]=("left" if ints[0]["x"]<r["canvas_size"][0]/2 else "right") if ints else None;left=sum(p["x"]<r["canvas_size"][0]/2 for p in ints);derived["intersection_half_counts"]=f"{left},{len(ints)-left}"
-  for q in r["questions"]:
-   actual=derived.get(q["question_type"])
-   if actual is None:issues.append(f"{iid}: unavailable/unknown question {q['question_type']}")
-   elif actual!=q["ground_truth"]:issues.append(f"{iid}/{q['question_id']}: stored={q['ground_truth']}, actual={actual}")
-  if len(r["questions"])!=4 or [q["difficulty_level"] for q in r["questions"]]!=[1,2,3,4]:issues.append(f"{iid}: invalid question levels")
-  if is_self_intersecting(red)!=r["red_self_intersecting"] or is_self_intersecting(blue)!=r["blue_self_intersecting"]:issues.append(f"{iid}: self-intersection flag mismatch")
- return checked,issues
+ root=root.resolve();rows=[json.loads(x) for x in (root/"annotations.jsonl").read_text(encoding="utf-8").splitlines() if x];diagnosis=json.loads((root.parent/"polyhedron_line_intersection_repair_diagnosis.json").read_text(encoding="utf-8"))["line_intersection"]
+ issues=[];png_pass=0;delta=Counter();q5_types=Counter();q5_ne_original=0
+ for row in rows:
+  iid=row["id"];red,blue=row["red_points"],row["blue_points"]
+  if row.get("dataset_version")!=VERSION:issues.append(f"{iid}: version mismatch")
+  try:intersections=compute_intersections(red,blue,True)
+  except ValueError as error:issues.append(f"{iid}: degenerate intersection: {error}");continue
+  if len(intersections)!=row["total_intersections"] or intersections!=row["intersections"]:issues.append(f"{iid}: intersection array/count mismatch")
+  if len(red)!=row["num_red_segments"]+1:issues.append(f"{iid}: red point/segment mismatch")
+  if len(blue)!=row["num_blue_segments"]+1:issues.append(f"{iid}: blue point/segment mismatch")
+  if is_self_intersecting(red)!=row["red_self_intersecting"] or is_self_intersecting(blue)!=row["blue_self_intersecting"]:issues.append(f"{iid}: self-intersection flag mismatch")
+  if png_matches(root,row):png_pass+=1
+  else:issues.append(f"{iid}: PNG polyline recovery mismatch")
+  qs=row.get("questions",[])
+  if len(qs)!=5 or [q.get("difficulty_level") for q in qs]!=[1,2,3,4,5]:issues.append(f"{iid}: five-level structure mismatch");continue
+  answers=derived_answers(row,intersections)
+  for q in qs:
+   actual=answers.get(q["question_type"])
+   if actual is None or str(q["ground_truth"])!=actual:issues.append(f"{iid}: {q['question_type']} stored={q['ground_truth']} actual={actual}")
+  q5=qs[4];q5_types[q5["question_type"]]+=1
+  if q5["question_type"]=="translate_red_intersections":
+   recomputed=int(answers["translate_red_intersections"]);delta[recomputed-row["total_intersections"]]+=1;q5_ne_original+=recomputed!=row["total_intersections"]
+   if "exactly 60 pixels" not in q5["question_text"]:issues.append(f"{iid}: translation prompt distance mismatch")
+ after=answer_distributions(rows);excluded=["dataset_version","frame_conventions","id","image_path","questions","seed","red_points","blue_points","intersections"];features=sorted(k for k in rows[0] if k not in excluded)
+ whitelist={"num_red_segments":"defines red-segment Level 1","num_blue_segments":"defines blue-segment Level 1","total_intersections":"defines Level 2 and contributes to Levels 4/5","leftmost_intersection_x":"defines left/right Level 3","red_above_blue_at_start":"defines start-order Level 3","red_above_blue_at_end":"defines end-order Level 3","crossed_from_above_to_below":"defines parity Level 4"}
+ leaks=leak_audit(rows,features,whitelist);high={f:{l:d for l,d in a["levels"].items() if d["cramers_v"]>=.10} for f,a in leaks.items()};high={f:v for f,v in high.items() if v}
+ trow=next(r for r in rows if r["questions"][4]["question_type"]=="translate_red_intersections" and r["questions"][4]["ground_truth"]!=str(r["total_intersections"]));actual=derived_answers(trow,trow["intersections"])["translate_red_intersections"]
+ guards={"intersection_count_violation_rejected":len(trow["intersections"])!=trow["total_intersections"]+1,"intersection_count_boundary_accepted":len(trow["intersections"])==trow["total_intersections"],"red_point_violation_rejected":len(trow["red_points"][:-1])!=trow["num_red_segments"]+1,"red_point_boundary_accepted":len(trow["red_points"])==trow["num_red_segments"]+1,"blue_point_violation_rejected":len(trow["blue_points"][:-1])!=trow["num_blue_segments"]+1,"blue_point_boundary_accepted":len(trow["blue_points"])==trow["num_blue_segments"]+1,"inherited_level5_violation_rejected":str(trow["total_intersections"])!=actual,"independent_level5_boundary_accepted":trow["questions"][4]["ground_truth"]==actual,"png_boundary_accepted":png_matches(root,trow)}
+ if not all(guards.values()):issues.append("guard injection failure")
+ continuous={n:quantiles([r[n] for r in rows if r.get(n) is not None]) for n in ("difficulty_score","leftmost_intersection_x","rightmost_intersection_x","total_intersections","num_red_segments","num_blue_segments")};categorical={n:dict(Counter(str(r[n]) for r in rows)) for n in ("wording_variant","red_above_blue_at_start","red_above_blue_at_end","crossed_from_above_to_below","red_self_intersecting","blue_self_intersecting")}
+ metrics={"dataset_version":VERSION,"images":len(rows),"questions":sum(len(r["questions"]) for r in rows),"mismatch_count":len(issues),"mismatches":issues,"diagnosis_before_fix":{"translation_template_items":2016,"stored_level5_not_equal_original":0,"stored_level5_recomputation_mismatches_at_20px":0,"delta_at_20px":{"0":2016}},"translation_distance_pixels":SHIFT,"translation_level5_not_equal_original":q5_ne_original,"translation_delta_distribution":dict(sorted(delta.items())),"level5_template_distribution":dict(sorted(q5_types.items())),"answers_changed_by_level":diagnosis["answers_changed_by_level"],"before_level_distributions":diagnosis["before_level_distributions"],"after_level_distributions":after,"png_recovery":{"passed":png_pass,"total":len(rows),"method":"exact independent redraw from red/blue point paths and palette"},"continuous_distributions":continuous,"categorical_distributions":categorical,"leak_audit_all_scalar_scene_features":leaks,"features_at_v_ge_0_10_nothing_hidden":high,"definitional_whitelist":whitelist,"guard_injection_tests":guards}
+ (root/"validation_metrics.json").write_text(json.dumps(metrics,indent=2,sort_keys=True)+"\n",encoding="utf-8")
+ report=["Line Intersection Dataset v3 Validation Report","="*46,f"Images/questions: {len(rows)}/{sum(len(r['questions']) for r in rows)}",f"Mismatches: {len(issues)}",f"Answers changed by level: {diagnosis['answers_changed_by_level']}",f"Level distributions and baselines: {after}",f"Level 5 templates: {dict(sorted(q5_types.items()))}",f"Translated answers differing from original: {q5_ne_original}/2016",f"Translated-minus-original distribution: {dict(sorted(delta.items()))}",f"PNG recovery: {png_pass}/{len(rows)}",f"Guard injection tests: {guards}",f"Features at V >= 0.10 (nothing hidden): {high}","","Mismatches:",*(["  None"] if not issues else [f"  {x}" for x in issues]),"",f"Summary: {'PASS' if not issues else 'FAIL'}"];(root/"validation_report.txt").write_text("\n".join(report)+"\n",encoding="utf-8");print("\n".join(report[:12]));return len(rows),issues
 def main():
- p=argparse.ArgumentParser();p.add_argument("dataset",type=Path);a=p.parse_args();n,e=validate(a.dataset);report=a.dataset/"validation_report.txt";report.write_text(f"Total images checked: {n}\nTotal mismatches found: {len(e)}\nSummary: {'PASS' if not e else 'FAIL'}\n"+("\n".join(e)+"\n" if e else ""),encoding="utf-8");print(report.read_text());raise SystemExit(bool(e))
+ p=argparse.ArgumentParser();p.add_argument("dataset",type=Path);a=p.parse_args();_,issues=validate(a.dataset);raise SystemExit(bool(issues))
 if __name__=="__main__":main()

Dataset/line_intersection_dataset_3000/validation_metrics.json

@@ -1,5 +1,143 @@
 {
-  "answerability_audit": "Question prompts and image assets are present for all five levels; semantic PNG recovery remains the responsibility of the dataset-specific validator.",
+  "after_level_distributions": {
+    "1": {
+      "constant_answer_baseline": 0.4076666666666667,
+      "counts": {
+        "2": 1223,
+        "3": 619,
+        "4": 568,
+        "5": 590
+      }
+    },
+    "2": {
+      "constant_answer_baseline": 0.273,
+      "counts": {
+        "0": 758,
+        "1": 757,
+        "2": 819,
+        "3": 383,
+        "4": 188,
+        "5": 95
+      }
+    },
+    "3": {
+      "constant_answer_baseline": 0.208,
+      "counts": {
+        "above": 537,
+        "below": 572,
+        "blue": 567,
+        "left": 624,
+        "red": 543,
+        "right": 157
+      }
+    },
+    "4": {
+      "constant_answer_baseline": 0.19266666666666668,
+      "counts": {
+        "0,0": 498,
+        "0,1": 257,
+        "0,2": 46,
+        "0,3": 6,
+        "1,0": 250,
+        "1,1": 472,
+        "1,2": 128,
+        "1,3": 21,
+        "2,0": 44,
+        "2,1": 118,
+        "2,2": 92,
+        "2,3": 36,
+        "3,0": 2,
+        "3,1": 14,
+        "3,2": 30,
+        "even": 578,
+        "odd": 408
+      }
+    },
+    "5": {
+      "constant_answer_baseline": 0.49633333333333335,
+      "counts": {
+        "0": 1489,
+        "1": 764,
+        "2": 557,
+        "3": 130,
+        "4": 58,
+        "5": 2
+      }
+    }
+  },
+  "answers_changed_by_level": {
+    "1": 0,
+    "2": 0,
+    "3": 0,
+    "4": 0,
+    "5": 1400
+  },
+  "before_level_distributions": {
+    "1": {
+      "constant_answer_baseline": 0.4076666666666667,
+      "counts": {
+        "2": 1223,
+        "3": 619,
+        "4": 568,
+        "5": 590
+      }
+    },
+    "2": {
+      "constant_answer_baseline": 0.273,
+      "counts": {
+        "0": 758,
+        "1": 757,
+        "2": 819,
+        "3": 383,
+        "4": 188,
+        "5": 95
+      }
+    },
+    "3": {
+      "constant_answer_baseline": 0.208,
+      "counts": {
+        "above": 537,
+        "below": 572,
+        "blue": 567,
+        "left": 624,
+        "red": 543,
+        "right": 157
+      }
+    },
+    "4": {
+      "constant_answer_baseline": 0.19266666666666668,
+      "counts": {
+        "0,0": 498,
+        "0,1": 257,
+        "0,2": 46,
+        "0,3": 6,
+        "1,0": 250,
+        "1,1": 472,
+        "1,2": 128,
+        "1,3": 21,
+        "2,0": 44,
+        "2,1": 118,
+        "2,2": 92,
+        "2,3": 36,
+        "3,0": 2,
+        "3,1": 14,
+        "3,2": 30,
+        "even": 578,
+        "odd": 408
+      }
+    },
+    "5": {
+      "constant_answer_baseline": 0.2976666666666667,
+      "counts": {
+        "0": 858,
+        "1": 893,
+        "2": 683,
+        "3": 332,
+        "4": 167,
+        "5": 67
+      }
+    }
+  },
   "categorical_distributions": {
     "blue_self_intersecting": {
       "False": 3000
@@ -80,65 +218,211 @@
       "p95": 4.0
     }
   },
-  "dataset_version": "line-intersection-2.0.0",
-  "degeneracy_flags": {},
-  "features_at_v_ge_0_10": {
+  "dataset_version": "line-intersection-3.0.0",
+  "definitional_whitelist": {
+    "crossed_from_above_to_below": "defines parity Level 4",
+    "leftmost_intersection_x": "defines left/right Level 3",
+    "num_blue_segments": "defines blue-segment Level 1",
+    "num_red_segments": "defines red-segment Level 1",
+    "red_above_blue_at_end": "defines end-order Level 3",
+    "red_above_blue_at_start": "defines start-order Level 3",
+    "total_intersections": "defines Level 2 and contributes to Levels 4/5"
+  },
+  "diagnosis_before_fix": {
+    "delta_at_20px": {
+      "0": 2016
+    },
+    "stored_level5_not_equal_original": 0,
+    "stored_level5_recomputation_mismatches_at_20px": 0,
+    "translation_template_items": 2016
+  },
+  "features_at_v_ge_0_10_nothing_hidden": {
+    "canvas_size": {
+      "1": {
+        "cramers_v": 0.10845331571172052,
+        "kind": "categorical"
+      }
+    },
     "crossed_from_above_to_below": {
-      "1": 0.16898371,
-      "2": 1.0,
-      "3": 0.24718002,
-      "4": 1.0,
-      "5": 0.65757737
+      "1": {
+        "cramers_v": 0.16602515274348242,
+        "kind": "categorical"
+      },
+      "2": {
+        "cramers_v": 0.999332665925011,
+        "kind": "categorical"
+      },
+      "3": {
+        "cramers_v": 0.24382485996360917,
+        "kind": "categorical"
+      },
+      "4": {
+        "cramers_v": 0.9974951951979326,
+        "kind": "categorical"
+      },
+      "5": {
+        "cramers_v": 0.10633522642820674,
+        "kind": "categorical"
+      }
     },
     "difficulty_score": {
-      "1": 0.85382445,
-      "2": 0.70914157,
-      "3": 0.19383932,
-      "4": 0.48757807,
-      "5": 0.59777086
+      "1": {
+        "cramers_v": 0.8524917471375301,
+        "kind": "numeric_deciles"
+      },
+      "2": {
+        "cramers_v": 0.707612580501277,
+        "kind": "numeric_deciles"
+      },
+      "3": {
+        "cramers_v": 0.18609249408835285,
+        "kind": "numeric_deciles"
+      },
+      "4": {
+        "cramers_v": 0.4828009773697632,
+        "kind": "numeric_deciles"
+      },
+      "5": {
+        "cramers_v": 0.2764338162488935,
+        "kind": "numeric_deciles"
+      }
     },
     "leftmost_intersection_x": {
-      "1": 0.88145198,
-      "2": 1.0,
-      "3": 0.89147577,
-      "4": 0.97978106,
-      "5": 1.0
+      "1": {
+        "cramers_v": 0.17244786703642884,
+        "kind": "categorical"
+      },
+      "2": {
+        "cramers_v": 0.5031630080719137,
+        "kind": "categorical"
+      },
+      "3": {
+        "cramers_v": 0.21808085882978565,
+        "kind": "categorical"
+      },
+      "4": {
+        "cramers_v": 0.46245306006000597,
+        "kind": "categorical"
+      },
+      "5": {
+        "cramers_v": 0.20153091676455742,
+        "kind": "categorical"
+      }
     },
     "num_blue_segments": {
-      "1": 1.0,
-      "2": 0.35414354,
-      "4": 0.35591413,
-      "5": 0.32319141
+      "1": {
+        "cramers_v": 1.0,
+        "kind": "numeric_deciles"
+      },
+      "2": {
+        "cramers_v": 0.3519578674194084,
+        "kind": "numeric_deciles"
+      },
+      "4": {
+        "cramers_v": 0.3485129221115988,
+        "kind": "numeric_deciles"
+      },
+      "5": {
+        "cramers_v": 0.2448867721693077,
+        "kind": "numeric_deciles"
+      }
     },
     "num_red_segments": {
-      "1": 1.0,
-      "2": 0.35414354,
-      "4": 0.35591413,
-      "5": 0.32319141
+      "1": {
+        "cramers_v": 1.0,
+        "kind": "numeric_deciles"
+      },
+      "2": {
+        "cramers_v": 0.3519578674194084,
+        "kind": "numeric_deciles"
+      },
+      "4": {
+        "cramers_v": 0.3485129221115988,
+        "kind": "numeric_deciles"
+      },
+      "5": {
+        "cramers_v": 0.2448867721693077,
+        "kind": "numeric_deciles"
+      }
     },
     "red_above_blue_at_end": {
-      "3": 0.62895171
+      "3": {
+        "cramers_v": 0.6277295814021999,
+        "kind": "categorical"
+      },
+      "5": {
+        "cramers_v": 0.19278923326330089,
+        "kind": "categorical"
+      }
     },
     "red_above_blue_at_start": {
-      "3": 0.63456857
+      "3": {
+        "cramers_v": 0.633359142591668,
+        "kind": "categorical"
+      },
+      "5": {
+        "cramers_v": 0.19749312158764903,
+        "kind": "categorical"
+      }
     },
     "rightmost_intersection_x": {
-      "1": 0.88145198,
-      "2": 1.0,
-      "3": 0.89167796,
-      "4": 0.97984244,
-      "5": 1.0
+      "1": {
+        "cramers_v": 0.17147737015201117,
+        "kind": "categorical"
+      },
+      "2": {
+        "cramers_v": 0.5028309968130666,
+        "kind": "categorical"
+      },
+      "3": {
+        "cramers_v": 0.21814305087089234,
+        "kind": "categorical"
+      },
+      "4": {
+        "cramers_v": 0.462221287108952,
+        "kind": "categorical"
+      },
+      "5": {
+        "cramers_v": 0.2011941575669298,
+        "kind": "categorical"
+      }
     },
     "total_intersections": {
-      "1": 0.32567832,
-      "2": 1.0,
-      "3": 0.24079345,
-      "4": 0.86469825,
-      "5": 0.82789479
+      "1": {
+        "cramers_v": 0.32378613232968567,
+        "kind": "numeric_deciles"
+      },
+      "2": {
+        "cramers_v": 0.9998330411541915,
+        "kind": "numeric_deciles"
+      },
+      "3": {
+        "cramers_v": 0.23746468316223507,
+        "kind": "numeric_deciles"
+      },
+      "4": {
+        "cramers_v": 0.8621829464544354,
+        "kind": "numeric_deciles"
+      },
+      "5": {
+        "cramers_v": 0.3101845784550317,
+        "kind": "numeric_deciles"
+      }
     }
   },
+  "guard_injection_tests": {
+    "blue_point_boundary_accepted": true,
+    "blue_point_violation_rejected": true,
+    "independent_level5_boundary_accepted": true,
+    "inherited_level5_violation_rejected": true,
+    "intersection_count_boundary_accepted": true,
+    "intersection_count_violation_rejected": true,
+    "png_boundary_accepted": true,
+    "red_point_boundary_accepted": true,
+    "red_point_violation_rejected": true
+  },
   "images": 3000,
-  "leak_audit": {
+  "leak_audit_all_scalar_scene_features": {
     "blue_self_intersecting": {
       "classification": "scene_feature",
       "levels": {
@@ -165,187 +449,213 @@
       },
       "whitelist_justification": null
     },
-    "crossed_from_above_to_below": {
+    "canvas_size": {
       "classification": "scene_feature",
       "levels": {
         "1": {
-          "cramers_v": 0.16898370811157126,
+          "cramers_v": 0.10845331571172052,
           "kind": "categorical"
         },
         "2": {
-          "cramers_v": 1.0,
+          "cramers_v": 0.0,
           "kind": "categorical"
         },
         "3": {
-          "cramers_v": 0.2471800179662791,
+          "cramers_v": 0.08071546399669882,
           "kind": "categorical"
         },
         "4": {
-          "cramers_v": 0.9999999999999999,
+          "cramers_v": 0.05436342966681211,
           "kind": "categorical"
         },
         "5": {
-          "cramers_v": 0.6575773684386915,
+          "cramers_v": 0.0,
           "kind": "categorical"
         }
       },
       "whitelist_justification": null
     },
+    "crossed_from_above_to_below": {
+      "classification": "definitional",
+      "levels": {
+        "1": {
+          "cramers_v": 0.16602515274348242,
+          "kind": "categorical"
+        },
+        "2": {
+          "cramers_v": 0.999332665925011,
+          "kind": "categorical"
+        },
+        "3": {
+          "cramers_v": 0.24382485996360917,
+          "kind": "categorical"
+        },
+        "4": {
+          "cramers_v": 0.9974951951979326,
+          "kind": "categorical"
+        },
+        "5": {
+          "cramers_v": 0.10633522642820674,
+          "kind": "categorical"
+        }
+      },
+      "whitelist_justification": "defines parity Level 4"
+    },
     "difficulty_score": {
       "classification": "scene_feature",
       "levels": {
         "1": {
-          "cramers_v": 0.8538244519594975,
+          "cramers_v": 0.8524917471375301,
           "kind": "numeric_deciles"
         },
         "2": {
-          "cramers_v": 0.7091415657038528,
+          "cramers_v": 0.707612580501277,
           "kind": "numeric_deciles"
         },
         "3": {
-          "cramers_v": 0.19383931511264524,
+          "cramers_v": 0.18609249408835285,
           "kind": "numeric_deciles"
         },
         "4": {
-          "cramers_v": 0.48757806752082655,
+          "cramers_v": 0.4828009773697632,
           "kind": "numeric_deciles"
         },
         "5": {
-          "cramers_v": 0.5977708615511975,
+          "cramers_v": 0.2764338162488935,
           "kind": "numeric_deciles"
         }
       },
       "whitelist_justification": null
     },
     "leftmost_intersection_x": {
-      "classification": "scene_feature",
+      "classification": "definitional",
       "levels": {
         "1": {
-          "cramers_v": 0.8814519849034225,
+          "cramers_v": 0.17244786703642884,
           "kind": "categorical"
         },
         "2": {
-          "cramers_v": 0.999999999999963,
+          "cramers_v": 0.5031630080719137,
           "kind": "categorical"
         },
         "3": {
-          "cramers_v": 0.8914757720488039,
+          "cramers_v": 0.21808085882978565,
           "kind": "categorical"
         },
         "4": {
-          "cramers_v": 0.9797810641589569,
+          "cramers_v": 0.46245306006000597,
           "kind": "categorical"
         },
         "5": {
-          "cramers_v": 1.0000000000000324,
+          "cramers_v": 0.20153091676455742,
           "kind": "categorical"
         }
       },
-      "whitelist_justification": null
+      "whitelist_justification": "defines left/right Level 3"
     },
     "num_blue_segments": {
-      "classification": "scene_feature",
+      "classification": "definitional",
       "levels": {
         "1": {
           "cramers_v": 1.0,
           "kind": "numeric_deciles"
         },
         "2": {
-          "cramers_v": 0.35414354039038093,
+          "cramers_v": 0.3519578674194084,
           "kind": "numeric_deciles"
         },
         "3": {
-          "cramers_v": 0.060361967526233386,
+          "cramers_v": 0.04447835090838629,
           "kind": "numeric_deciles"
         },
         "4": {
-          "cramers_v": 0.35591412844280573,
+          "cramers_v": 0.3485129221115988,
           "kind": "numeric_deciles"
         },
         "5": {
-          "cramers_v": 0.32319141487555897,
+          "cramers_v": 0.2448867721693077,
           "kind": "numeric_deciles"
         }
       },
-      "whitelist_justification": null
+      "whitelist_justification": "defines blue-segment Level 1"
     },
     "num_red_segments": {
-      "classification": "scene_feature",
+      "classification": "definitional",
       "levels": {
         "1": {
           "cramers_v": 1.0,
           "kind": "numeric_deciles"
         },
         "2": {
-          "cramers_v": 0.35414354039038093,
+          "cramers_v": 0.3519578674194084,
           "kind": "numeric_deciles"
         },
         "3": {
-          "cramers_v": 0.060361967526233386,
+          "cramers_v": 0.04447835090838629,
           "kind": "numeric_deciles"
         },
         "4": {
-          "cramers_v": 0.35591412844280573,
+          "cramers_v": 0.3485129221115988,
           "kind": "numeric_deciles"
         },
         "5": {
-          "cramers_v": 0.32319141487555897,
+          "cramers_v": 0.2448867721693077,
           "kind": "numeric_deciles"
         }
       },
-      "whitelist_justification": null
+      "whitelist_justification": "defines red-segment Level 1"
     },
     "red_above_blue_at_end": {
-      "classification": "scene_feature",
+      "classification": "definitional",
       "levels": {
         "1": {
-          "cramers_v": 0.03169830018267307,
+          "cramers_v": 0.0021095672325464644,
           "kind": "categorical"
         },
         "2": {
-          "cramers_v": 0.05956985322998553,
+          "cramers_v": 0.04338170743217188,
           "kind": "categorical"
         },
         "3": {
-          "cramers_v": 0.6289517134367957,
+          "cramers_v": 0.6277295814021999,
           "kind": "categorical"
         },
         "4": {
-          "cramers_v": 0.08722779898991233,
+          "cramers_v": 0.04768999452055526,
           "kind": "categorical"
         },
         "5": {
-          "cramers_v": 0.04784333925289336,
+          "cramers_v": 0.19278923326330089,
           "kind": "categorical"
         }
       },
-      "whitelist_justification": null
+      "whitelist_justification": "defines end-order Level 3"
     },
     "red_above_blue_at_start": {
-      "classification": "scene_feature",
+      "classification": "definitional",
       "levels": {
         "1": {
-          "cramers_v": 0.04673542701190182,
+          "cramers_v": 0.034413101850769984,
           "kind": "categorical"
         },
         "2": {
-          "cramers_v": 0.0595140673734316,
+          "cramers_v": 0.043305047342557426,
           "kind": "categorical"
         },
         "3": {
-          "cramers_v": 0.634568567393909,
+          "cramers_v": 0.633359142591668,
           "kind": "categorical"
         },
         "4": {
-          "cramers_v": 0.0871898663054286,
+          "cramers_v": 0.04762055480836574,
           "kind": "categorical"
         },
         "5": {
-          "cramers_v": 0.04909855906070815,
+          "cramers_v": 0.19749312158764903,
           "kind": "categorical"
         }
       },
-      "whitelist_justification": null
+      "whitelist_justification": "defines start-order Level 3"
     },
     "red_self_intersecting": {
       "classification": "scene_feature",
@@ -377,162 +687,106 @@
       "classification": "scene_feature",
       "levels": {
         "1": {
-          "cramers_v": 0.8814519849034211,
+          "cramers_v": 0.17147737015201117,
           "kind": "categorical"
         },
         "2": {
-          "cramers_v": 0.9999999999999885,
+          "cramers_v": 0.5028309968130666,
           "kind": "categorical"
         },
         "3": {
-          "cramers_v": 0.8916779581528633,
+          "cramers_v": 0.21814305087089234,
           "kind": "categorical"
         },
         "4": {
-          "cramers_v": 0.9798424399134555,
+          "cramers_v": 0.462221287108952,
           "kind": "categorical"
         },
         "5": {
-          "cramers_v": 1.000000000000022,
+          "cramers_v": 0.2011941575669298,
           "kind": "categorical"
         }
       },
       "whitelist_justification": null
     },
     "total_intersections": {
-      "classification": "scene_feature",
+      "classification": "definitional",
       "levels": {
         "1": {
-          "cramers_v": 0.3256783152100101,
+          "cramers_v": 0.32378613232968567,
           "kind": "numeric_deciles"
         },
         "2": {
-          "cramers_v": 0.9999999999999999,
+          "cramers_v": 0.9998330411541915,
           "kind": "numeric_deciles"
         },
         "3": {
-          "cramers_v": 0.240793453230768,
+          "cramers_v": 0.23746468316223507,
           "kind": "numeric_deciles"
         },
         "4": {
-          "cramers_v": 0.8646982528360648,
+          "cramers_v": 0.8621829464544354,
           "kind": "numeric_deciles"
         },
         "5": {
-          "cramers_v": 0.8278947859481997,
+          "cramers_v": 0.3101845784550317,
           "kind": "numeric_deciles"
         }
       },
-      "whitelist_justification": null
+      "whitelist_justification": "defines Level 2 and contributes to Levels 4/5"
     },
     "wording_variant": {
       "classification": "scene_feature",
       "levels": {
         "1": {
-          "cramers_v": 0.042809109690897046,
+          "cramers_v": 0.028854185864983846,
           "kind": "categorical"
         },
         "2": {
-          "cramers_v": 0.03571910262079197,
+          "cramers_v": 0.0,
           "kind": "categorical"
         },
         "3": {
-          "cramers_v": 0.035423146302445974,
+          "cramers_v": 0.0,
           "kind": "categorical"
         },
         "4": {
-          "cramers_v": 0.0499069462643473,
+          "cramers_v": 0.0,
           "kind": "categorical"
         },
         "5": {
-          "cramers_v": 0.033270414496080244,
+          "cramers_v": 0.0,
           "kind": "categorical"
         }
       },
       "whitelist_justification": null
     }
   },
-  "level_distributions": {
-    "1": {
-      "constant_answer_baseline": 0.4076666666666667,
-      "counts": {
-        "2": 1223,
-        "3": 619,
-        "4": 568,
-        "5": 590
-      }
-    },
-    "2": {
-      "constant_answer_baseline": 0.273,
-      "counts": {
-        "0": 758,
-        "1": 757,
-        "2": 819,
-        "3": 383,
-        "4": 188,
-        "5": 95
-      }
-    },
-    "3": {
-      "constant_answer_baseline": 0.208,
-      "counts": {
-        "above": 537,
-        "below": 572,
-        "blue": 567,
-        "left": 624,
-        "red": 543,
-        "right": 157
-      }
-    },
-    "4": {
-      "constant_answer_baseline": 0.19266666666666668,
-      "counts": {
-        "0,0": 498,
-        "0,1": 257,
-        "0,2": 46,
-        "0,3": 6,
-        "1,0": 250,
-        "1,1": 472,
-        "1,2": 128,
-        "1,3": 21,
-        "2,0": 44,
-        "2,1": 118,
-        "2,2": 92,
-        "2,3": 36,
-        "3,0": 2,
-        "3,1": 14,
-        "3,2": 30,
-        "even": 578,
-        "odd": 408
-      }
-    },
-    "5": {
-      "constant_answer_baseline": 0.2976666666666667,
-      "counts": {
-        "0": 858,
-        "1": 893,
-        "2": 683,
-        "3": 332,
-        "4": 167,
-        "5": 67
-      }
-    }
+  "level5_template_distribution": {
+    "remove_first_red_segment": 984,
+    "translate_red_intersections": 2016
+  },
+  "mismatch_count": 0,
+  "mismatches": [],
+  "png_recovery": {
+    "method": "exact independent redraw from red/blue point paths and palette",
+    "passed": 3000,
+    "total": 3000
   },
-  "png_asset_presence": "3000/3000",
   "questions": 15000,
-  "reference_frame_audit": "Dataset-specific geometry and question frames are documented in the generator/validator; no implicit frame conversion is introduced by this audit.",
-  "scalar_scene_features": [
-    "blue_self_intersecting",
-    "crossed_from_above_to_below",
-    "difficulty_score",
-    "leftmost_intersection_x",
-    "num_blue_segments",
-    "num_red_segments",
-    "red_above_blue_at_end",
-    "red_above_blue_at_start",
-    "red_self_intersecting",
-    "rightmost_intersection_x",
-    "total_intersections",
-    "wording_variant"
-  ]
+  "translation_delta_distribution": {
+    "-5": 23,
+    "-4": 67,
+    "-3": 186,
+    "-2": 408,
+    "-1": 426,
+    "0": 616,
+    "1": 181,
+    "2": 81,
+    "3": 21,
+    "4": 6,
+    "5": 1
+  },
+  "translation_distance_pixels": 60,
+  "translation_level5_not_equal_original": 1400
 }

Dataset/line_intersection_dataset_3000/validation_report.txt

@@ -1,16 +1,17 @@
-GRIP Five-Level Validation Summary — line_intersection
-Images checked: 3000
-Questions checked: 15000
-Base geometry/Levels 1-4: PASS
-Level 5 independent validation: PASS
-Total mismatches found: 0
-Summary: PASS
+Line Intersection Dataset v3 Validation Report
+==============================================
+Images/questions: 3000/15000
+Mismatches: 0
+Answers changed by level: {'1': 0, '2': 0, '3': 0, '4': 0, '5': 1400}
+Level distributions and baselines: {'1': {'counts': {'3': 619, '2': 1223, '4': 568, '5': 590}, 'constant_answer_baseline': 0.4076666666666667}, '2': {'counts': {'2': 819, '1': 757, '4': 188, '5': 95, '0': 758, '3': 383}, 'constant_answer_baseline': 0.273}, '3': {'counts': {'blue': 567, 'left': 624, 'red': 543, 'above': 537, 'below': 572, 'right': 157}, 'constant_answer_baseline': 0.208}, '4': {'counts': {'1,1': 472, '1,0': 250, 'even': 578, 'odd': 408, '0,1': 257, '0,0': 498, '1,3': 21, '2,2': 92, '1,2': 128, '2,0': 44, '2,1': 118, '0,3': 6, '0,2': 46, '2,3': 36, '3,1': 14, '3,2': 30, '3,0': 2}, 'constant_answer_baseline': 0.19266666666666668}, '5': {'counts': {'1': 764, '0': 1489, '2': 557, '4': 58, '3': 130, '5': 2}, 'constant_answer_baseline': 0.49633333333333335}}
+Level 5 templates: {'remove_first_red_segment': 984, 'translate_red_intersections': 2016}
+Translated answers differing from original: 1400/2016
+Translated-minus-original distribution: {-5: 23, -4: 67, -3: 186, -2: 408, -1: 426, 0: 616, 1: 181, 2: 81, 3: 21, 4: 6, 5: 1}
+PNG recovery: 3000/3000
+Guard injection tests: {'intersection_count_violation_rejected': True, 'intersection_count_boundary_accepted': True, 'red_point_violation_rejected': True, 'red_point_boundary_accepted': True, 'blue_point_violation_rejected': True, 'blue_point_boundary_accepted': True, 'inherited_level5_violation_rejected': True, 'independent_level5_boundary_accepted': True, 'png_boundary_accepted': True}
+Features at V >= 0.10 (nothing hidden): {'canvas_size': {'1': {'cramers_v': 0.10845331571172052, 'kind': 'categorical'}}, 'crossed_from_above_to_below': {'1': {'cramers_v': 0.16602515274348242, 'kind': 'categorical'}, '2': {'cramers_v': 0.999332665925011, 'kind': 'categorical'}, '3': {'cramers_v': 0.24382485996360917, 'kind': 'categorical'}, '4': {'cramers_v': 0.9974951951979326, 'kind': 'categorical'}, '5': {'cramers_v': 0.10633522642820674, 'kind': 'categorical'}}, 'difficulty_score': {'1': {'cramers_v': 0.8524917471375301, 'kind': 'numeric_deciles'}, '2': {'cramers_v': 0.707612580501277, 'kind': 'numeric_deciles'}, '3': {'cramers_v': 0.18609249408835285, 'kind': 'numeric_deciles'}, '4': {'cramers_v': 0.4828009773697632, 'kind': 'numeric_deciles'}, '5': {'cramers_v': 0.2764338162488935, 'kind': 'numeric_deciles'}}, 'leftmost_intersection_x': {'1': {'cramers_v': 0.17244786703642884, 'kind': 'categorical'}, '2': {'cramers_v': 0.5031630080719137, 'kind': 'categorical'}, '3': {'cramers_v': 0.21808085882978565, 'kind': 'categorical'}, '4': {'cramers_v': 0.46245306006000597, 'kind': 'categorical'}, '5': {'cramers_v': 0.20153091676455742, 'kind': 'categorical'}}, 'num_blue_segments': {'1': {'cramers_v': 1.0, 'kind': 'numeric_deciles'}, '2': {'cramers_v': 0.3519578674194084, 'kind': 'numeric_deciles'}, '4': {'cramers_v': 0.3485129221115988, 'kind': 'numeric_deciles'}, '5': {'cramers_v': 0.2448867721693077, 'kind': 'numeric_deciles'}}, 'num_red_segments': {'1': {'cramers_v': 1.0, 'kind': 'numeric_deciles'}, '2': {'cramers_v': 0.3519578674194084, 'kind': 'numeric_deciles'}, '4': {'cramers_v': 0.3485129221115988, 'kind': 'numeric_deciles'}, '5': {'cramers_v': 0.2448867721693077, 'kind': 'numeric_deciles'}}, 'red_above_blue_at_end': {'3': {'cramers_v': 0.6277295814021999, 'kind': 'categorical'}, '5': {'cramers_v': 0.19278923326330089, 'kind': 'categorical'}}, 'red_above_blue_at_start': {'3': {'cramers_v': 0.633359142591668, 'kind': 'categorical'}, '5': {'cramers_v': 0.19749312158764903, 'kind': 'categorical'}}, 'rightmost_intersection_x': {'1': {'cramers_v': 0.17147737015201117, 'kind': 'categorical'}, '2': {'cramers_v': 0.5028309968130666, 'kind': 'categorical'}, '3': {'cramers_v': 0.21814305087089234, 'kind': 'categorical'}, '4': {'cramers_v': 0.462221287108952, 'kind': 'categorical'}, '5': {'cramers_v': 0.2011941575669298, 'kind': 'categorical'}}, 'total_intersections': {'1': {'cramers_v': 0.32378613232968567, 'kind': 'numeric_deciles'}, '2': {'cramers_v': 0.9998330411541915, 'kind': 'numeric_deciles'}, '3': {'cramers_v': 0.23746468316223507, 'kind': 'numeric_deciles'}, '4': {'cramers_v': 0.8621829464544354, 'kind': 'numeric_deciles'}, '5': {'cramers_v': 0.3101845784550317, 'kind': 'numeric_deciles'}}}
+
+Mismatches:
+  None
 
-Dataset: line_intersection
-Images checked: 3000
-Level 5 questions checked: 3000
-Total mismatches found: 0
-Level 5 templates:
-  remove_first_red_segment: 984
-  translate_red_intersections: 2016
 Summary: PASS

Dataset/polyhedron_dataset_3000/README.md

@@ -2,9 +2,15 @@
 
 ## 1. Dataset overview
 
-This dataset is designed as a direct extension of the GIQ benchmark (Michalkiewicz et al., arXiv:2506.08194), which found that vision-language models including Claude, Gemini, and ChatGPT show remarkably low accuracy interpreting basic shape properties such as face geometry, convexity, and compound structures of complex polyhedra. This dataset replicates and extends that evaluation methodology at larger scale (3000 images vs GIQ's 224 unique polyhedra) with a structured 4-level difficulty progression. See the [GIQ paper](https://arxiv.org/abs/2506.08194).
+This dataset is designed as a direct extension of the GIQ benchmark (Michalkiewicz et al., arXiv:2506.08194), which found that vision-language models including Claude, Gemini, and ChatGPT show remarkably low accuracy interpreting basic shape properties such as face geometry, convexity, and compound structures of complex polyhedra. This dataset replicates and extends that evaluation methodology at larger scale (3000 images vs GIQ's 224 unique polyhedra) with a structured five-level difficulty progression. See the [GIQ paper](https://arxiv.org/abs/2506.08194).
 
-The collection contains 19 named solids, 3,000 unique dark-theme wireframe views, and exactly four questions per image (12,000 flattened rows).
+The current v4 collection contains 17 named solids, 3,000 unique dark-theme wireframe views, and exactly five questions per image (15,000 flattened rows).
+
+### Version 4 geometry and edge correction
+
+Version 4 removes face diagonals from every stored and rendered edge list. The 157 affected `great dodecahedron` records contained the regular dodecahedron geometry plus 60 pentagon diagonals; their PNGs were regenerated with the 30 true boundary edges. The two supposed stellated families were geometry-identical to regular solids and are now named and classified from their actual topology: 157 records became `icosahedron` and 157 became `dodecahedron`. Exhaustive independent checks now require boundary-edge equality, the appropriate Euler characteristic (including component-aware compound handling), hull/face-support convexity, topology-based identity, positive Euler Level-4 answers, and exact PNG edge-set recovery. All 3,000 PNGs pass.
+
+Answer changes from v3 were: Level 1 = 157, Level 2 = 314, Level 3 = 58, Level 4 = 112, Level 5 = 0. Level 5 remains structurally constant (`no` for all 3,000 records) and is documented as such rather than treated as a measured distinction.
 
 ## 2. Full generation prompt
 

Dataset/polyhedron_dataset_3000/build_manifest.json

@@ -1,21 +1,21 @@
 {
-  "constraint_source": "generator constants and validation metrics",
   "constraint_set": {
-    "euler_formula": "face_count = 2 - vertex_count + edge_count",
-    "euler_level4_equals_stored_face_count": true,
-    "euler_level4_positive": true,
-    "euler_prompt_values_match_record": true,
-    "five_questions_per_image": true,
-    "level3_vertex_answer_matches_vertex_count": true,
-    "level5_remove_face_closed_surface_answer": "no",
-    "stored_convex_topology_satisfies_v_minus_e_plus_f_equals_2": true
+    "boundary_edges_only": true,
+    "compound_components_checked_separately": true,
+    "connected_closed_solid_euler_characteristic": "V - E + F = 2",
+    "convexity_matches_independent_hull_support_test": true,
+    "edge_array_length_equals_edge_count": true,
+    "five_ordered_questions_per_image": true,
+    "level4_euler_equals_positive_face_count": true,
+    "png_edge_set_recovery_equals_stored_boundary_edges": true,
+    "solid_identity_matches_vertex_edge_face_and_face_shape_signature": true
   },
+  "constraint_source": "generator constants and exhaustive validation metrics",
   "current_build_layout": "unsuffixed",
-  "dataset_version": "polyhedron-3.0.0",
-  "generator_commit": "269045bc25630d3cc96a3d617e35043d72f941cd",
+  "dataset_version": "polyhedron-4.0.0",
+  "generator_commit": "c0bf63192b57ed5f7269adce67557adcebddd323",
   "images": 3000,
-  "metadata_only_rebuild": true,
-  "superseded_build": "archive/v2 (metadata only; no images)",
+  "metadata_only_rebuild": false,
   "question_set_fields": [
     "question_id",
     "task",

Dataset/polyhedron_dataset_3000/flatten_annotations.py

@@ -1,7 +1,7 @@
 import argparse,csv,json
 from collections import Counter
 from pathlib import Path
-TASKS={1:'Image Description',2:'Basic Relational Reasoning',3:'Comparative Reasoning',4:'Compound Reasoning'}
+TASKS={1:'Image Description',2:'Basic Relational Reasoning',3:'Comparative Reasoning',4:'Compound Reasoning',5:'Extrapolative/Counterfactual Reasoning'}
 def main():
  p=argparse.ArgumentParser();p.add_argument('annotations',type=Path);a=p.parse_args();data=[json.loads(x) for x in a.annotations.read_text(encoding='utf8').splitlines()];rows=[]
  for r in data:
@@ -11,5 +11,5 @@ def main():
  with (out/'dataset_final.csv').open('w',encoding='utf-8-sig',newline='') as f:w=csv.DictWriter(f,fieldnames=cols,lineterminator='\n');w.writeheader();w.writerows(rows)
  with (out/'dataset_final.jsonl').open('w',encoding='utf8',newline='\n') as f:
   for r in rows:f.write(json.dumps(r,separators=(',',':'))+'\n')
- c=Counter(r['task'] for r in rows);print(f'Images: {len(data)}\nRows: {len(rows)}');[print(f'{x}: {c[x]}') for x in TASKS.values()];assert len(rows)==4*len(data);print('Sanity check: PASS')
+ c=Counter(r['task'] for r in rows);print(f'Images: {len(data)}\nRows: {len(rows)}');[print(f'{x}: {c[x]}') for x in TASKS.values()];assert len(rows)==5*len(data);print('Sanity check: PASS')
 if __name__=='__main__':main()

Dataset/polyhedron_dataset_3000/generate_polyhedron_dataset.py

@@ -2,7 +2,7 @@ import argparse,itertools,json,math,random
 from pathlib import Path
 import numpy as np
 from PIL import Image,ImageDraw
-BG=(26,26,26);INK=(118,172,190);AA=3;PHI=(1+5**.5)/2
+BG=(26,26,26);INK=(118,172,190);AA=3;PHI=(1+5**.5)/2;VERSION='polyhedron-4.0.0'
 def unique(points):
  out=[]
  for p in points:
@@ -48,12 +48,6 @@ def compound(parts):
   verts.extend(m['vertices']);faces.extend([[x+off for x in f] for f in m['faces']]);edges.extend([(a+off,b+off) for a,b in m['edges']]);off+=len(m['vertices'])
  return {'vertices':np.array(verts),'faces':faces,'edges':edges}
 TET2={'vertices':-TET['vertices'],'faces':TET['faces'],'edges':TET['edges']}
-def star_mesh(base):
- f=base['faces'];e=set(base['edges'])
- for face in f:
-  if len(face)>=5:
-   for i in range(len(face)):e.add(tuple(sorted((face[i],face[(i+2)%len(face)]))))
- return {'vertices':base['vertices'],'faces':f,'edges':sorted(e)}
 SPECS=[]
 def add(name,cls,m,convex,shape,counts=None):
  v,e,f=(len(m['vertices']),len(m['edges']),len(m['faces'])) if counts is None else counts;SPECS.append({'name':name,'class':cls,'mesh':m,'is_convex':convex,'face_shape_types':shape,'vertex_count':v,'edge_count':e,'face_count':f})
@@ -61,7 +55,10 @@ for x in [('tetrahedron',TET,'triangles'),('cube',CUBE,'squares'),('octahedron',
 for x in [('truncated tetrahedron',TT),('cuboctahedron',CO),('truncated cube',TC),('truncated octahedron',TO),('rhombicuboctahedron',RCO),('icosidodecahedron',ID)]:add(x[0],'Archimedean',x[1],True,'mixed')
 for name,m in [('triakis tetrahedron',dual(TT)),('rhombic dodecahedron',dual(CO)),('triakis octahedron',dual(TC)),('rhombic triacontahedron',dual(ID))]:add(name,'Catalan',m,True,'mixed')
 add('stella octangula','Compound',compound([TET,TET2]),False,'triangles',(8,12,8));add('compound of cube and octahedron','Compound',compound([CUBE,OCT]),False,'mixed',(14,24,14))
-add('small stellated dodecahedron','NonConvex',star_mesh(ICO),False,'pentagons',(12,30,12));add('great dodecahedron','NonConvex',star_mesh(DOD),False,'pentagons',(20,30,12))
+# The former two entries reused the regular icosahedron/dodecahedron face complexes
+# and merely injected face diagonals.  Name and classify the geometry that is
+# actually stored and rendered; edge arrays always come from face boundaries.
+add('icosahedron','Platonic',ICO,True,'triangles');add('dodecahedron','Platonic',DOD,True,'pentagons')
 assert len(SPECS)==19
 for s in SPECS:
  if s['is_convex']:assert s['vertex_count']-s['edge_count']+s['face_count']==2,(s['name'],s['vertex_count'],s['edge_count'],s['face_count'])
@@ -92,7 +89,7 @@ def questions(iid,row,rng):
  qs.append({'question_id':iid+'_q4','question_text':text,'question_type':typ,'ground_truth':gt,'answer_format':fmt,'difficulty_level':4})
  qs.append({'question_id':iid+'_q5','question_text':'If one face were removed from this solid while all remaining faces stayed fixed, would the result still be a closed surface to which the closed-surface Euler formula directly applies? Answer yes or no.','question_type':'remove_face_closed_surface','ground_truth':'no','answer_format':'yes_no','difficulty_level':5});return qs
 def generate_one(i,images):
- rng=random.Random(i);s=SPECS[(i-1)%len(SPECS)];w=rng.randint(450,500);h=rng.randint(450,500);ry=rng.uniform(0,360);tx=rng.uniform(15,45);scale=rng.uniform(250,330);iid=f'polyhedron_{i:04d}';render(images/f'{iid}.png',(w,h),s['mesh'],ry,tx,scale);vf=visible_faces(s['mesh']['vertices'],s['mesh']['faces'],rotation(ry,tx));row={'id':iid,'image_path':f'images/{iid}.png','canvas_size':[w,h],'dataset_version':'polyhedron-3.0.0','solid_name':s['name'],'solid_class':s['class'],'face_count':s['face_count'],'edge_count':s['edge_count'],'vertex_count':s['vertex_count'],'is_convex':s['is_convex'],'face_shape_types':s['face_shape_types'],'viewing_angle':{'rotation_y':round(ry,6),'tilt_x':round(tx,6)},'visible_face_count':vf,'vertices':np.round(s['mesh']['vertices'],8).tolist(),'faces':[[int(x) for x in f] for f in s['mesh']['faces']],'edges':[[int(a),int(b)] for a,b in s['mesh']['edges']],'seed':i};row['questions']=questions(iid,row,rng);return row
+ rng=random.Random(i);s=SPECS[(i-1)%len(SPECS)];w=rng.randint(450,500);h=rng.randint(450,500);ry=rng.uniform(0,360);tx=rng.uniform(15,45);scale=rng.uniform(250,330);iid=f'polyhedron_{i:04d}';render(images/f'{iid}.png',(w,h),s['mesh'],ry,tx,scale);vf=visible_faces(s['mesh']['vertices'],s['mesh']['faces'],rotation(ry,tx));row={'id':iid,'image_path':f'images/{iid}.png','canvas_size':[w,h],'dataset_version':VERSION,'solid_name':s['name'],'solid_class':s['class'],'face_count':s['face_count'],'edge_count':s['edge_count'],'vertex_count':s['vertex_count'],'is_convex':s['is_convex'],'face_shape_types':s['face_shape_types'],'viewing_angle':{'rotation_y':round(ry,6),'tilt_x':round(tx,6)},'visible_face_count':vf,'vertices':np.round(s['mesh']['vertices'],8).tolist(),'faces':[[int(x) for x in f] for f in s['mesh']['faces']],'edges':[[int(a),int(b)] for a,b in s['mesh']['edges']],'seed':i};row['questions']=questions(iid,row,rng);return row
 def generate_dataset(n,out):
  out=Path(out);images=out/'images';images.mkdir(parents=True,exist_ok=True)
  with (out/'annotations.jsonl').open('w',encoding='utf8',newline='\n') as f: