diff --git a/models/abl_full/rf/seed42/outer_fold_0/test_metrics.json b/models/abl_full/rf/seed42/outer_fold_0/test_metrics.json
index 2bd2ce7..cdc207e 100644
--- a/models/abl_full/rf/seed42/outer_fold_0/test_metrics.json
+++ b/models/abl_full/rf/seed42/outer_fold_0/test_metrics.json
@@ -1,42 +1,105 @@
 {
   "size": {
     "n_samples": 83,
-    "mse": 1.368858521000182,
-    "rmse": 1.1699822737974204,
-    "mae": 0.48265016079386586,
-    "r2": 0.16222489685168928
+    "mse": 1.3484356700772522,
+    "rmse": 1.1612216283196124,
+    "mae": 0.4798589242061937,
+    "r2": 0.1747241783006852
   },
   "delivery": {
     "n_samples": 58,
-    "mse": 0.4106486248647463,
-    "rmse": 0.6408187145088275,
-    "mae": 0.4092221012826776,
-    "r2": 0.4780408138229023
+    "mse": 0.3976989005700695,
+    "rmse": 0.6306337293311146,
+    "mae": 0.4073375633664443,
+    "r2": 0.49450069866587976
   },
   "toxic": {
     "n_samples": 58,
-    "accuracy": 1.0,
-    "precision": 1.0,
-    "recall": 1.0,
-    "f1": 1.0
+    "accuracy": 0.9655172413793104,
+    "precision": 0.75,
+    "recall": 0.9821428571428572,
+    "f1": 0.8242424242424242,
+    "true_class_counts": [
+      56,
+      2
+    ],
+    "pred_class_counts": [
+      54,
+      4
+    ],
+    "confusion_matrix": [
+      [
+        54,
+        2
+      ],
+      [
+        0,
+        2
+      ]
+    ]
   },
   "pdi": {
     "n_samples": 84,
-    "accuracy": 0.8095238095238095,
-    "precision": 0.7797101449275362,
-    "recall": 0.7063435495367071,
-    "f1": 0.7279352226720648
+    "accuracy": 0.7857142857142857,
+    "precision": 0.7551020408163265,
+    "recall": 0.8118317890235209,
+    "f1": 0.7630094043887148,
+    "true_class_counts": [
+      61,
+      23
+    ],
+    "pred_class_counts": [
+      49,
+      35
+    ],
+    "confusion_matrix": [
+      [
+        46,
+        15
+      ],
+      [
+        3,
+        20
+      ]
+    ]
   },
   "ee": {
     "n_samples": 84,
     "accuracy": 0.75,
-    "precision": 0.6912280701754385,
-    "recall": 0.6471428571428571,
-    "f1": 0.6631611379274931
+    "precision": 0.7092916445857623,
+    "recall": 0.7614285714285715,
+    "f1": 0.7175824990730441,
+    "true_class_counts": [
+      14,
+      20,
+      50
+    ],
+    "pred_class_counts": [
+      17,
+      30,
+      37
+    ],
+    "confusion_matrix": [
+      [
+        10,
+        3,
+        1
+      ],
+      [
+        3,
+        17,
+        0
+      ],
+      [
+        4,
+        10,
+        36
+      ]
+    ]
   },
   "biodist": {
     "n_samples": 58,
-    "kl_divergence": 0.16366098804686893,
-    "js_divergence": 0.0359968761663274
+    "kl_divergence": 0.15998570647485624,
+    "js_divergence": 0.035472610690609384
   }
 }
\ No newline at end of file
diff --git a/models/abl_full/rf/seed42/outer_fold_1/test_metrics.json b/models/abl_full/rf/seed42/outer_fold_1/test_metrics.json
index 613583a..ec4465f 100644
--- a/models/abl_full/rf/seed42/outer_fold_1/test_metrics.json
+++ b/models/abl_full/rf/seed42/outer_fold_1/test_metrics.json
@@ -4,39 +4,102 @@
     "mse": 0.30493822479834903,
     "rmse": 0.5522121193874225,
     "mae": 0.34891908283267253,
-    "r2": 0.0688270762246912
+    "r2": 0.06882707391945686
   },
   "delivery": {
     "n_samples": 61,
-    "mse": 0.7776284206466513,
-    "rmse": 0.8818324220886026,
-    "mae": 0.519796750021438,
-    "r2": 0.4113913613868899
+    "mse": 0.7752937492958416,
+    "rmse": 0.8805076656655759,
+    "mae": 0.5151286462071908,
+    "r2": 0.4131585218812348
   },
   "toxic": {
     "n_samples": 61,
-    "accuracy": 1.0,
-    "precision": 1.0,
-    "recall": 1.0,
-    "f1": 1.0
+    "accuracy": 0.9508196721311475,
+    "precision": 0.75,
+    "recall": 0.9741379310344828,
+    "f1": 0.8200589970501475,
+    "true_class_counts": [
+      58,
+      3
+    ],
+    "pred_class_counts": [
+      55,
+      6
+    ],
+    "confusion_matrix": [
+      [
+        55,
+        3
+      ],
+      [
+        0,
+        3
+      ]
+    ]
   },
   "pdi": {
     "n_samples": 84,
-    "accuracy": 0.7619047619047619,
-    "precision": 0.6711111111111111,
-    "recall": 0.5873015873015873,
-    "f1": 0.5942028985507246
+    "accuracy": 0.6547619047619048,
+    "precision": 0.5803571428571429,
+    "recall": 0.5952380952380952,
+    "f1": 0.5822328931572629,
+    "true_class_counts": [
+      63,
+      21
+    ],
+    "pred_class_counts": [
+      56,
+      28
+    ],
+    "confusion_matrix": [
+      [
+        45,
+        18
+      ],
+      [
+        11,
+        10
+      ]
+    ]
   },
   "ee": {
     "n_samples": 84,
-    "accuracy": 0.6785714285714286,
-    "precision": 0.6062341913405743,
-    "recall": 0.6071785357499643,
-    "f1": 0.6041666666666666
+    "accuracy": 0.5476190476190477,
+    "precision": 0.5001165151969107,
+    "recall": 0.4989534275248561,
+    "f1": 0.48349530196256724,
+    "true_class_counts": [
+      13,
+      22,
+      49
+    ],
+    "pred_class_counts": [
+      19,
+      31,
+      34
+    ],
+    "confusion_matrix": [
+      [
+        5,
+        7,
+        1
+      ],
+      [
+        8,
+        11,
+        3
+      ],
+      [
+        6,
+        13,
+        30
+      ]
+    ]
   },
   "biodist": {
     "n_samples": 61,
     "kl_divergence": 0.1177531067102451,
-    "js_divergence": 0.027774748827162937
+    "js_divergence": 0.027774748827162944
   }
 }
\ No newline at end of file
diff --git a/models/abl_full/rf/seed42/outer_fold_2/test_metrics.json b/models/abl_full/rf/seed42/outer_fold_2/test_metrics.json
index c33e2c9..355fdb3 100644
--- a/models/abl_full/rf/seed42/outer_fold_2/test_metrics.json
+++ b/models/abl_full/rf/seed42/outer_fold_2/test_metrics.json
@@ -1,42 +1,105 @@
 {
   "size": {
     "n_samples": 84,
-    "mse": 0.3089202253953697,
-    "rmse": 0.5558059242175902,
-    "mae": 0.3994691472061615,
-    "r2": 0.3898880719888205
+    "mse": 0.3091245824692921,
+    "rmse": 0.5559897323416073,
+    "mae": 0.4046506446335209,
+    "r2": 0.3894844727418765
   },
   "delivery": {
     "n_samples": 60,
-    "mse": 0.6215887874238653,
-    "rmse": 0.7884090229213928,
-    "mae": 0.4654098044336172,
-    "r2": 0.17255759426449502
+    "mse": 0.623471858388545,
+    "rmse": 0.789602341934562,
+    "mae": 0.4688382081680016,
+    "r2": 0.17005089391015027
   },
   "toxic": {
     "n_samples": 61,
-    "accuracy": 1.0,
-    "precision": 1.0,
-    "recall": 1.0,
-    "f1": 1.0
+    "accuracy": 0.9344262295081968,
+    "precision": 0.7142857142857143,
+    "recall": 0.9655172413793103,
+    "f1": 0.7821428571428571,
+    "true_class_counts": [
+      58,
+      3
+    ],
+    "pred_class_counts": [
+      54,
+      7
+    ],
+    "confusion_matrix": [
+      [
+        54,
+        4
+      ],
+      [
+        0,
+        3
+      ]
+    ]
   },
   "pdi": {
     "n_samples": 84,
-    "accuracy": 0.7261904761904762,
-    "precision": 0.6428571428571428,
-    "recall": 0.6385630498533724,
-    "f1": 0.6405581395348837
+    "accuracy": 0.6904761904761905,
+    "precision": 0.6693363844393593,
+    "recall": 0.717008797653959,
+    "f1": 0.662962962962963,
+    "true_class_counts": [
+      62,
+      22
+    ],
+    "pred_class_counts": [
+      46,
+      38
+    ],
+    "confusion_matrix": [
+      [
+        41,
+        21
+      ],
+      [
+        5,
+        17
+      ]
+    ]
   },
   "ee": {
     "n_samples": 84,
-    "accuracy": 0.6547619047619048,
-    "precision": 0.5676343636869953,
-    "recall": 0.5378151260504201,
-    "f1": 0.5491025070464323
+    "accuracy": 0.6428571428571429,
+    "precision": 0.5842305037957212,
+    "recall": 0.6349206349206349,
+    "f1": 0.5925185925185925,
+    "true_class_counts": [
+      12,
+      21,
+      51
+    ],
+    "pred_class_counts": [
+      21,
+      23,
+      40
+    ],
+    "confusion_matrix": [
+      [
+        8,
+        2,
+        2
+      ],
+      [
+        5,
+        12,
+        4
+      ],
+      [
+        8,
+        9,
+        34
+      ]
+    ]
   },
   "biodist": {
     "n_samples": 60,
-    "kl_divergence": 0.11869463263891707,
-    "js_divergence": 0.02929511833384903
+    "kl_divergence": 0.11799692057231032,
+    "js_divergence": 0.029125166530752162
   }
 }
\ No newline at end of file
diff --git a/models/abl_full/rf/seed42/outer_fold_3/test_metrics.json b/models/abl_full/rf/seed42/outer_fold_3/test_metrics.json
index ff98c84..767f18a 100644
--- a/models/abl_full/rf/seed42/outer_fold_3/test_metrics.json
+++ b/models/abl_full/rf/seed42/outer_fold_3/test_metrics.json
@@ -1,42 +1,105 @@
 {
   "size": {
     "n_samples": 83,
-    "mse": 1.6827545971608378,
-    "rmse": 1.2972103133882484,
-    "mae": 0.5269114175759023,
-    "r2": 0.048221123148072476
+    "mse": 1.6732277890580922,
+    "rmse": 1.2935330645399414,
+    "mae": 0.5211672051452704,
+    "r2": 0.05360960932568781
   },
   "delivery": {
     "n_samples": 59,
-    "mse": 0.8527625246024667,
+    "mse": 0.8527625246024666,
     "rmse": 0.9234514197306032,
     "mae": 0.5525073854104011,
-    "r2": 0.1710134699931437
+    "r2": 0.17101350859831832
   },
   "toxic": {
     "n_samples": 60,
-    "accuracy": 1.0,
-    "precision": 1.0,
-    "recall": 1.0,
-    "f1": 1.0
+    "accuracy": 0.9333333333333333,
+    "precision": 0.7142857142857143,
+    "recall": 0.9649122807017544,
+    "f1": 0.7818181818181817,
+    "true_class_counts": [
+      57,
+      3
+    ],
+    "pred_class_counts": [
+      53,
+      7
+    ],
+    "confusion_matrix": [
+      [
+        53,
+        4
+      ],
+      [
+        0,
+        3
+      ]
+    ]
   },
   "pdi": {
     "n_samples": 84,
-    "accuracy": 0.75,
-    "precision": 0.6676909569798068,
-    "recall": 0.6400293255131965,
-    "f1": 0.6493738819320214
+    "accuracy": 0.7380952380952381,
+    "precision": 0.6612903225806452,
+    "recall": 0.6612903225806452,
+    "f1": 0.6612903225806452,
+    "true_class_counts": [
+      62,
+      22
+    ],
+    "pred_class_counts": [
+      62,
+      22
+    ],
+    "confusion_matrix": [
+      [
+        51,
+        11
+      ],
+      [
+        11,
+        11
+      ]
+    ]
   },
   "ee": {
     "n_samples": 84,
-    "accuracy": 0.6309523809523809,
-    "precision": 0.5132964178288214,
-    "recall": 0.49489693313222727,
-    "f1": 0.4975934917795383
+    "accuracy": 0.6666666666666666,
+    "precision": 0.5872039589961117,
+    "recall": 0.6381096028154852,
+    "f1": 0.5938808373590982,
+    "true_class_counts": [
+      13,
+      20,
+      51
+    ],
+    "pred_class_counts": [
+      23,
+      20,
+      41
+    ],
+    "confusion_matrix": [
+      [
+        10,
+        3,
+        0
+      ],
+      [
+        9,
+        8,
+        3
+      ],
+      [
+        4,
+        9,
+        38
+      ]
+    ]
   },
   "biodist": {
     "n_samples": 60,
-    "kl_divergence": 0.1458100103551393,
-    "js_divergence": 0.034398655034422854
+    "kl_divergence": 0.14886390502458902,
+    "js_divergence": 0.03503301624445018
   }
 }
\ No newline at end of file
diff --git a/models/abl_full/rf/seed42/outer_fold_4/test_metrics.json b/models/abl_full/rf/seed42/outer_fold_4/test_metrics.json
index bcec505..8d7caf0 100644
--- a/models/abl_full/rf/seed42/outer_fold_4/test_metrics.json
+++ b/models/abl_full/rf/seed42/outer_fold_4/test_metrics.json
@@ -1,42 +1,105 @@
 {
   "size": {
     "n_samples": 84,
-    "mse": 0.539048457673242,
-    "rmse": 0.7341991948192548,
-    "mae": 0.41399459761950447,
-    "r2": 0.27868624280784415
+    "mse": 0.5439429374081691,
+    "rmse": 0.7375248723996833,
+    "mae": 0.41855190159962563,
+    "r2": 0.2721367833807379
   },
   "delivery": {
     "n_samples": 58,
-    "mse": 0.8685551446715326,
-    "rmse": 0.9319630597140278,
-    "mae": 0.48785093074168856,
-    "r2": 0.16012146519578374
+    "mse": 0.8807361070482976,
+    "rmse": 0.9384754163260206,
+    "mae": 0.4923390269539427,
+    "r2": 0.14834272376574187
   },
   "toxic": {
     "n_samples": 59,
-    "accuracy": 1.0,
-    "precision": 1.0,
-    "recall": 1.0,
-    "f1": 1.0
+    "accuracy": 0.9661016949152542,
+    "precision": 0.8,
+    "recall": 0.9821428571428572,
+    "f1": 0.865909090909091,
+    "true_class_counts": [
+      56,
+      3
+    ],
+    "pred_class_counts": [
+      54,
+      5
+    ],
+    "confusion_matrix": [
+      [
+        54,
+        2
+      ],
+      [
+        0,
+        3
+      ]
+    ]
   },
   "pdi": {
     "n_samples": 84,
-    "accuracy": 0.7380952380952381,
-    "precision": 0.6515151515151515,
-    "recall": 0.6319648093841642,
-    "f1": 0.6390625
+    "accuracy": 0.6785714285714286,
+    "precision": 0.6428571428571428,
+    "recall": 0.6796187683284457,
+    "f1": 0.6415362731152205,
+    "true_class_counts": [
+      62,
+      22
+    ],
+    "pred_class_counts": [
+      49,
+      35
+    ],
+    "confusion_matrix": [
+      [
+        42,
+        20
+      ],
+      [
+        7,
+        15
+      ]
+    ]
   },
   "ee": {
     "n_samples": 84,
-    "accuracy": 0.6785714285714286,
-    "precision": 0.6048840048840048,
-    "recall": 0.6175213675213675,
-    "f1": 0.6107062011796588
+    "accuracy": 0.6309523809523809,
+    "precision": 0.5673525820584645,
+    "recall": 0.6021367521367521,
+    "f1": 0.5701075744179193,
+    "true_class_counts": [
+      12,
+      20,
+      52
+    ],
+    "pred_class_counts": [
+      17,
+      28,
+      39
+    ],
+    "confusion_matrix": [
+      [
+        7,
+        4,
+        1
+      ],
+      [
+        6,
+        11,
+        3
+      ],
+      [
+        4,
+        13,
+        35
+      ]
+    ]
   },
   "biodist": {
     "n_samples": 58,
-    "kl_divergence": 0.12253709805850461,
-    "js_divergence": 0.03003058068886421
+    "kl_divergence": 0.12072341813237816,
+    "js_divergence": 0.029660750070876794
   }
 }
\ No newline at end of file
diff --git a/models/abl_full/rf/seed42/summary.json b/models/abl_full/rf/seed42/summary.json
index 546d1b3..6e3659f 100644
--- a/models/abl_full/rf/seed42/summary.json
+++ b/models/abl_full/rf/seed42/summary.json
@@ -5,43 +5,106 @@
       "test_metrics": {
         "size": {
           "n_samples": 83,
-          "mse": 1.368858521000182,
-          "rmse": 1.1699822737974204,
-          "mae": 0.48265016079386586,
-          "r2": 0.16222489685168928
+          "mse": 1.3484356700772522,
+          "rmse": 1.1612216283196124,
+          "mae": 0.4798589242061937,
+          "r2": 0.1747241783006852
         },
         "delivery": {
           "n_samples": 58,
-          "mse": 0.4106486248647463,
-          "rmse": 0.6408187145088275,
-          "mae": 0.4092221012826776,
-          "r2": 0.4780408138229023
+          "mse": 0.3976989005700695,
+          "rmse": 0.6306337293311146,
+          "mae": 0.4073375633664443,
+          "r2": 0.49450069866587976
         },
         "toxic": {
           "n_samples": 58,
-          "accuracy": 1.0,
-          "precision": 1.0,
-          "recall": 1.0,
-          "f1": 1.0
+          "accuracy": 0.9655172413793104,
+          "precision": 0.75,
+          "recall": 0.9821428571428572,
+          "f1": 0.8242424242424242,
+          "true_class_counts": [
+            56,
+            2
+          ],
+          "pred_class_counts": [
+            54,
+            4
+          ],
+          "confusion_matrix": [
+            [
+              54,
+              2
+            ],
+            [
+              0,
+              2
+            ]
+          ]
         },
         "pdi": {
           "n_samples": 84,
-          "accuracy": 0.8095238095238095,
-          "precision": 0.7797101449275362,
-          "recall": 0.7063435495367071,
-          "f1": 0.7279352226720648
+          "accuracy": 0.7857142857142857,
+          "precision": 0.7551020408163265,
+          "recall": 0.8118317890235209,
+          "f1": 0.7630094043887148,
+          "true_class_counts": [
+            61,
+            23
+          ],
+          "pred_class_counts": [
+            49,
+            35
+          ],
+          "confusion_matrix": [
+            [
+              46,
+              15
+            ],
+            [
+              3,
+              20
+            ]
+          ]
         },
         "ee": {
           "n_samples": 84,
           "accuracy": 0.75,
-          "precision": 0.6912280701754385,
-          "recall": 0.6471428571428571,
-          "f1": 0.6631611379274931
+          "precision": 0.7092916445857623,
+          "recall": 0.7614285714285715,
+          "f1": 0.7175824990730441,
+          "true_class_counts": [
+            14,
+            20,
+            50
+          ],
+          "pred_class_counts": [
+            17,
+            30,
+            37
+          ],
+          "confusion_matrix": [
+            [
+              10,
+              3,
+              1
+            ],
+            [
+              3,
+              17,
+              0
+            ],
+            [
+              4,
+              10,
+              36
+            ]
+          ]
         },
         "biodist": {
           "n_samples": 58,
-          "kl_divergence": 0.16366098804686893,
-          "js_divergence": 0.0359968761663274
+          "kl_divergence": 0.15998570647485624,
+          "js_divergence": 0.035472610690609384
         }
       }
     },
@@ -53,40 +116,103 @@
           "mse": 0.30493822479834903,
           "rmse": 0.5522121193874225,
           "mae": 0.34891908283267253,
-          "r2": 0.0688270762246912
+          "r2": 0.06882707391945686
         },
         "delivery": {
           "n_samples": 61,
-          "mse": 0.7776284206466513,
-          "rmse": 0.8818324220886026,
-          "mae": 0.519796750021438,
-          "r2": 0.4113913613868899
+          "mse": 0.7752937492958416,
+          "rmse": 0.8805076656655759,
+          "mae": 0.5151286462071908,
+          "r2": 0.4131585218812348
         },
         "toxic": {
           "n_samples": 61,
-          "accuracy": 1.0,
-          "precision": 1.0,
-          "recall": 1.0,
-          "f1": 1.0
+          "accuracy": 0.9508196721311475,
+          "precision": 0.75,
+          "recall": 0.9741379310344828,
+          "f1": 0.8200589970501475,
+          "true_class_counts": [
+            58,
+            3
+          ],
+          "pred_class_counts": [
+            55,
+            6
+          ],
+          "confusion_matrix": [
+            [
+              55,
+              3
+            ],
+            [
+              0,
+              3
+            ]
+          ]
         },
         "pdi": {
           "n_samples": 84,
-          "accuracy": 0.7619047619047619,
-          "precision": 0.6711111111111111,
-          "recall": 0.5873015873015873,
-          "f1": 0.5942028985507246
+          "accuracy": 0.6547619047619048,
+          "precision": 0.5803571428571429,
+          "recall": 0.5952380952380952,
+          "f1": 0.5822328931572629,
+          "true_class_counts": [
+            63,
+            21
+          ],
+          "pred_class_counts": [
+            56,
+            28
+          ],
+          "confusion_matrix": [
+            [
+              45,
+              18
+            ],
+            [
+              11,
+              10
+            ]
+          ]
         },
         "ee": {
           "n_samples": 84,
-          "accuracy": 0.6785714285714286,
-          "precision": 0.6062341913405743,
-          "recall": 0.6071785357499643,
-          "f1": 0.6041666666666666
+          "accuracy": 0.5476190476190477,
+          "precision": 0.5001165151969107,
+          "recall": 0.4989534275248561,
+          "f1": 0.48349530196256724,
+          "true_class_counts": [
+            13,
+            22,
+            49
+          ],
+          "pred_class_counts": [
+            19,
+            31,
+            34
+          ],
+          "confusion_matrix": [
+            [
+              5,
+              7,
+              1
+            ],
+            [
+              8,
+              11,
+              3
+            ],
+            [
+              6,
+              13,
+              30
+            ]
+          ]
         },
         "biodist": {
           "n_samples": 61,
           "kl_divergence": 0.1177531067102451,
-          "js_divergence": 0.027774748827162937
+          "js_divergence": 0.027774748827162944
         }
       }
     },
@@ -95,43 +221,106 @@
       "test_metrics": {
         "size": {
           "n_samples": 84,
-          "mse": 0.3089202253953697,
-          "rmse": 0.5558059242175902,
-          "mae": 0.3994691472061615,
-          "r2": 0.3898880719888205
+          "mse": 0.3091245824692921,
+          "rmse": 0.5559897323416073,
+          "mae": 0.4046506446335209,
+          "r2": 0.3894844727418765
         },
         "delivery": {
           "n_samples": 60,
-          "mse": 0.6215887874238653,
-          "rmse": 0.7884090229213928,
-          "mae": 0.4654098044336172,
-          "r2": 0.17255759426449502
+          "mse": 0.623471858388545,
+          "rmse": 0.789602341934562,
+          "mae": 0.4688382081680016,
+          "r2": 0.17005089391015027
         },
         "toxic": {
           "n_samples": 61,
-          "accuracy": 1.0,
-          "precision": 1.0,
-          "recall": 1.0,
-          "f1": 1.0
+          "accuracy": 0.9344262295081968,
+          "precision": 0.7142857142857143,
+          "recall": 0.9655172413793103,
+          "f1": 0.7821428571428571,
+          "true_class_counts": [
+            58,
+            3
+          ],
+          "pred_class_counts": [
+            54,
+            7
+          ],
+          "confusion_matrix": [
+            [
+              54,
+              4
+            ],
+            [
+              0,
+              3
+            ]
+          ]
         },
         "pdi": {
           "n_samples": 84,
-          "accuracy": 0.7261904761904762,
-          "precision": 0.6428571428571428,
-          "recall": 0.6385630498533724,
-          "f1": 0.6405581395348837
+          "accuracy": 0.6904761904761905,
+          "precision": 0.6693363844393593,
+          "recall": 0.717008797653959,
+          "f1": 0.662962962962963,
+          "true_class_counts": [
+            62,
+            22
+          ],
+          "pred_class_counts": [
+            46,
+            38
+          ],
+          "confusion_matrix": [
+            [
+              41,
+              21
+            ],
+            [
+              5,
+              17
+            ]
+          ]
         },
         "ee": {
           "n_samples": 84,
-          "accuracy": 0.6547619047619048,
-          "precision": 0.5676343636869953,
-          "recall": 0.5378151260504201,
-          "f1": 0.5491025070464323
+          "accuracy": 0.6428571428571429,
+          "precision": 0.5842305037957212,
+          "recall": 0.6349206349206349,
+          "f1": 0.5925185925185925,
+          "true_class_counts": [
+            12,
+            21,
+            51
+          ],
+          "pred_class_counts": [
+            21,
+            23,
+            40
+          ],
+          "confusion_matrix": [
+            [
+              8,
+              2,
+              2
+            ],
+            [
+              5,
+              12,
+              4
+            ],
+            [
+              8,
+              9,
+              34
+            ]
+          ]
         },
         "biodist": {
           "n_samples": 60,
-          "kl_divergence": 0.11869463263891707,
-          "js_divergence": 0.02929511833384903
+          "kl_divergence": 0.11799692057231032,
+          "js_divergence": 0.029125166530752162
         }
       }
     },
@@ -140,43 +329,106 @@
       "test_metrics": {
         "size": {
           "n_samples": 83,
-          "mse": 1.6827545971608378,
-          "rmse": 1.2972103133882484,
-          "mae": 0.5269114175759023,
-          "r2": 0.048221123148072476
+          "mse": 1.6732277890580922,
+          "rmse": 1.2935330645399414,
+          "mae": 0.5211672051452704,
+          "r2": 0.05360960932568781
         },
         "delivery": {
           "n_samples": 59,
-          "mse": 0.8527625246024667,
+          "mse": 0.8527625246024666,
           "rmse": 0.9234514197306032,
           "mae": 0.5525073854104011,
-          "r2": 0.1710134699931437
+          "r2": 0.17101350859831832
         },
         "toxic": {
           "n_samples": 60,
-          "accuracy": 1.0,
-          "precision": 1.0,
-          "recall": 1.0,
-          "f1": 1.0
+          "accuracy": 0.9333333333333333,
+          "precision": 0.7142857142857143,
+          "recall": 0.9649122807017544,
+          "f1": 0.7818181818181817,
+          "true_class_counts": [
+            57,
+            3
+          ],
+          "pred_class_counts": [
+            53,
+            7
+          ],
+          "confusion_matrix": [
+            [
+              53,
+              4
+            ],
+            [
+              0,
+              3
+            ]
+          ]
         },
         "pdi": {
           "n_samples": 84,
-          "accuracy": 0.75,
-          "precision": 0.6676909569798068,
-          "recall": 0.6400293255131965,
-          "f1": 0.6493738819320214
+          "accuracy": 0.7380952380952381,
+          "precision": 0.6612903225806452,
+          "recall": 0.6612903225806452,
+          "f1": 0.6612903225806452,
+          "true_class_counts": [
+            62,
+            22
+          ],
+          "pred_class_counts": [
+            62,
+            22
+          ],
+          "confusion_matrix": [
+            [
+              51,
+              11
+            ],
+            [
+              11,
+              11
+            ]
+          ]
         },
         "ee": {
           "n_samples": 84,
-          "accuracy": 0.6309523809523809,
-          "precision": 0.5132964178288214,
-          "recall": 0.49489693313222727,
-          "f1": 0.4975934917795383
+          "accuracy": 0.6666666666666666,
+          "precision": 0.5872039589961117,
+          "recall": 0.6381096028154852,
+          "f1": 0.5938808373590982,
+          "true_class_counts": [
+            13,
+            20,
+            51
+          ],
+          "pred_class_counts": [
+            23,
+            20,
+            41
+          ],
+          "confusion_matrix": [
+            [
+              10,
+              3,
+              0
+            ],
+            [
+              9,
+              8,
+              3
+            ],
+            [
+              4,
+              9,
+              38
+            ]
+          ]
         },
         "biodist": {
           "n_samples": 60,
-          "kl_divergence": 0.1458100103551393,
-          "js_divergence": 0.034398655034422854
+          "kl_divergence": 0.14886390502458902,
+          "js_divergence": 0.03503301624445018
         }
       }
     },
@@ -185,103 +437,166 @@
       "test_metrics": {
         "size": {
           "n_samples": 84,
-          "mse": 0.539048457673242,
-          "rmse": 0.7341991948192548,
-          "mae": 0.41399459761950447,
-          "r2": 0.27868624280784415
+          "mse": 0.5439429374081691,
+          "rmse": 0.7375248723996833,
+          "mae": 0.41855190159962563,
+          "r2": 0.2721367833807379
         },
         "delivery": {
           "n_samples": 58,
-          "mse": 0.8685551446715326,
-          "rmse": 0.9319630597140278,
-          "mae": 0.48785093074168856,
-          "r2": 0.16012146519578374
+          "mse": 0.8807361070482976,
+          "rmse": 0.9384754163260206,
+          "mae": 0.4923390269539427,
+          "r2": 0.14834272376574187
         },
         "toxic": {
           "n_samples": 59,
-          "accuracy": 1.0,
-          "precision": 1.0,
-          "recall": 1.0,
-          "f1": 1.0
+          "accuracy": 0.9661016949152542,
+          "precision": 0.8,
+          "recall": 0.9821428571428572,
+          "f1": 0.865909090909091,
+          "true_class_counts": [
+            56,
+            3
+          ],
+          "pred_class_counts": [
+            54,
+            5
+          ],
+          "confusion_matrix": [
+            [
+              54,
+              2
+            ],
+            [
+              0,
+              3
+            ]
+          ]
         },
         "pdi": {
           "n_samples": 84,
-          "accuracy": 0.7380952380952381,
-          "precision": 0.6515151515151515,
-          "recall": 0.6319648093841642,
-          "f1": 0.6390625
+          "accuracy": 0.6785714285714286,
+          "precision": 0.6428571428571428,
+          "recall": 0.6796187683284457,
+          "f1": 0.6415362731152205,
+          "true_class_counts": [
+            62,
+            22
+          ],
+          "pred_class_counts": [
+            49,
+            35
+          ],
+          "confusion_matrix": [
+            [
+              42,
+              20
+            ],
+            [
+              7,
+              15
+            ]
+          ]
         },
         "ee": {
           "n_samples": 84,
-          "accuracy": 0.6785714285714286,
-          "precision": 0.6048840048840048,
-          "recall": 0.6175213675213675,
-          "f1": 0.6107062011796588
+          "accuracy": 0.6309523809523809,
+          "precision": 0.5673525820584645,
+          "recall": 0.6021367521367521,
+          "f1": 0.5701075744179193,
+          "true_class_counts": [
+            12,
+            20,
+            52
+          ],
+          "pred_class_counts": [
+            17,
+            28,
+            39
+          ],
+          "confusion_matrix": [
+            [
+              7,
+              4,
+              1
+            ],
+            [
+              6,
+              11,
+              3
+            ],
+            [
+              4,
+              13,
+              35
+            ]
+          ]
         },
         "biodist": {
           "n_samples": 58,
-          "kl_divergence": 0.12253709805850461,
-          "js_divergence": 0.03003058068886421
+          "kl_divergence": 0.12072341813237816,
+          "js_divergence": 0.029660750070876794
         }
       }
     }
   ],
   "summary_stats": {
     "size": {
-      "mse_mean": 0.840904005205596,
-      "mse_std": 0.5742527703456151,
-      "rmse_mean": 0.8618819651219873,
-      "rmse_std": 0.3131508955808008,
-      "mae_mean": 0.4343888812056213,
-      "mae_std": 0.06297614136064734,
-      "r2_mean": 0.1895694822042235,
-      "r2_std": 0.12913183030644884
+      "mse_mean": 0.8359338407622309,
+      "mse_std": 0.5671793909058501,
+      "rmse_mean": 0.8600962833976535,
+      "rmse_std": 0.3101100192637678,
+      "mae_mean": 0.43462955168345657,
+      "mae_std": 0.06006325295257579,
+      "r2_mean": 0.19175642353368885,
+      "r2_std": 0.12652560874612837
     },
     "delivery": {
-      "mse_mean": 0.7062367004418524,
-      "mse_std": 0.17172737219125497,
-      "rmse_mean": 0.8332949277926908,
-      "rmse_std": 0.1088864718724354,
-      "mae_mean": 0.4869573943779645,
-      "mae_std": 0.048750584479951106,
-      "r2_mean": 0.27862494093264295,
-      "r2_std": 0.13730782732933916
+      "mse_mean": 0.705992627981044,
+      "mse_std": 0.1781763346919953,
+      "rmse_mean": 0.8325341145975752,
+      "rmse_std": 0.11348821970704938,
+      "mae_mean": 0.4872301660211962,
+      "mae_std": 0.048552308082167685,
+      "r2_mean": 0.279413269364265,
+      "r2_std": 0.14494172624677884
     },
     "toxic": {
-      "accuracy_mean": 1.0,
-      "accuracy_std": 0.0,
-      "precision_mean": 1.0,
-      "precision_std": 0.0,
-      "recall_mean": 1.0,
-      "recall_std": 0.0,
-      "f1_mean": 1.0,
-      "f1_std": 0.0
+      "accuracy_mean": 0.9500396342534485,
+      "accuracy_std": 0.014290090778642328,
+      "precision_mean": 0.7457142857142858,
+      "precision_std": 0.03149343955006944,
+      "recall_mean": 0.9737706334802525,
+      "recall_std": 0.007575117962473199,
+      "f1_mean": 0.8148343102325404,
+      "f1_std": 0.031251364844790935
     },
     "pdi": {
-      "accuracy_mean": 0.7571428571428571,
-      "accuracy_std": 0.02876915707998708,
-      "precision_mean": 0.6825769014781496,
-      "precision_std": 0.04965967017816626,
-      "recall_mean": 0.6408404643178055,
-      "recall_std": 0.03805702915009694,
-      "f1_mean": 0.6502265285379389,
-      "f1_std": 0.04335003701571735
+      "accuracy_mean": 0.7095238095238094,
+      "accuracy_std": 0.04677829215330594,
+      "precision_mean": 0.6617886067101234,
+      "precision_std": 0.05613216163532689,
+      "recall_mean": 0.6929975545649333,
+      "recall_std": 0.07132933119437145,
+      "f1_mean": 0.6622063712409613,
+      "f1_std": 0.05828494354685521
     },
     "ee": {
-      "accuracy_mean": 0.6785714285714286,
-      "accuracy_std": 0.03984095364447979,
-      "precision_mean": 0.5966554095831669,
-      "precision_std": 0.05812812328518918,
-      "recall_mean": 0.5809109639193673,
-      "recall_std": 0.055986847376187095,
-      "f1_mean": 0.5849460009199577,
-      "f1_std": 0.056684824199185184
+      "accuracy_mean": 0.6476190476190476,
+      "accuracy_std": 0.0650309537321317,
+      "precision_mean": 0.5896390409265941,
+      "precision_std": 0.06762084933247267,
+      "recall_mean": 0.62710979776526,
+      "recall_std": 0.08398957802940274,
+      "f1_mean": 0.5915169610662443,
+      "f1_std": 0.07486808628755862
     },
     "biodist": {
-      "kl_divergence_mean": 0.13369116716193502,
-      "kl_divergence_std": 0.01815702398766046,
-      "js_divergence_mean": 0.031499195810125294,
-      "js_divergence_std": 0.0031471347478191324
+      "kl_divergence_mean": 0.13306461138287579,
+      "kl_divergence_std": 0.017822152356421223,
+      "js_divergence_mean": 0.031413258472770295,
+      "js_divergence_std": 0.003197698848519836
     }
   }
 }
\ No newline at end of file
diff --git a/scripts/run_classical_baselines.py b/scripts/run_classical_baselines.py
index 5652b5c..ad1c858 100644
--- a/scripts/run_classical_baselines.py
+++ b/scripts/run_classical_baselines.py
@@ -13,10 +13,11 @@ import numpy as np
 import pandas as pd
 from scipy.special import rel_entr
 from sklearn.ensemble import RandomForestRegressor, RandomForestClassifier
-from sklearn.model_selection import KFold
+from sklearn.model_selection import KFold, StratifiedKFold
 from sklearn.preprocessing import StandardScaler
 from sklearn.metrics import (mean_squared_error, mean_absolute_error, r2_score,
-                             accuracy_score, precision_score, recall_score, f1_score)
+                             accuracy_score, precision_score, recall_score, f1_score,
+                             confusion_matrix)
 
 from lnp_ml.dataset import LNPDataset
 from lnp_ml.featurization.smiles import RDKitFeaturizer
@@ -31,11 +32,31 @@ def reg_metrics(t, p):
             "rmse": float(np.sqrt(mean_squared_error(t, p))),
             "mae": float(mean_absolute_error(t, p)), "r2": float(r2_score(t, p))}
 
-def clf_metrics(t, p):
-    return {"n_samples": int(len(p)), "accuracy": float(accuracy_score(t, p)),
-            "precision": float(precision_score(t, p, average="macro", zero_division=0)),
-            "recall": float(recall_score(t, p, average="macro", zero_division=0)),
-            "f1": float(f1_score(t, p, average="macro", zero_division=0))}
+def clf_metrics(t, p, n_classes):
+    """Classification metrics over the declared label space.
+
+    Supplying ``labels`` prevents a class absent from one test fold from being
+    silently dropped from the macro average.  Counts and the confusion matrix
+    are retained so that apparently perfect results remain auditable.
+    """
+    t, p = np.asarray(t, dtype=int), np.asarray(p, dtype=int)
+    labels = np.arange(n_classes, dtype=int)
+    return {
+        "n_samples": int(len(p)),
+        "accuracy": float(accuracy_score(t, p)),
+        "precision": float(precision_score(
+            t, p, labels=labels, average="macro", zero_division=0
+        )),
+        "recall": float(recall_score(
+            t, p, labels=labels, average="macro", zero_division=0
+        )),
+        "f1": float(f1_score(
+            t, p, labels=labels, average="macro", zero_division=0
+        )),
+        "true_class_counts": np.bincount(t, minlength=n_classes).astype(int).tolist(),
+        "pred_class_counts": np.bincount(p, minlength=n_classes).astype(int).tolist(),
+        "confusion_matrix": confusion_matrix(t, p, labels=labels).astype(int).tolist(),
+    }
 
 def dist_metrics(t, p, eps=1e-10):
     t = np.clip(np.asarray(t), eps, 1.0); p = np.clip(np.asarray(p), eps, 1.0)
@@ -44,11 +65,13 @@ def dist_metrics(t, p, eps=1e-10):
     js = float((0.5*np.sum(rel_entr(t, m), axis=-1) + 0.5*np.sum(rel_entr(p, m), axis=-1)).mean())
     return {"n_samples": int(len(p)), "kl_divergence": kl, "js_divergence": js}
 
-METRICS = {"reg": reg_metrics, "clf": clf_metrics, "dist": dist_metrics}
+METRICS = {"reg": reg_metrics, "dist": dist_metrics}
 
-def score(task_type, t, p):
+def score(task_type, t, p, n_classes=None):
     if task_type == "reg": return r2_score(t, p)
-    if task_type == "clf": return f1_score(t, p, average="macro", zero_division=0)
+    if task_type == "clf":
+        labels = np.arange(n_classes, dtype=int)
+        return f1_score(t, p, labels=labels, average="macro", zero_division=0)
     return -dist_metrics(t, p)["js_divergence"]
 
 # ---------- 相似度 / 距离 ----------
@@ -87,9 +110,26 @@ def knn_from_matrix(S, is_sim, ytr, k, task_type, n_classes):
     return np.array(out)
 
 # ---------- 各模型预测 ----------
-RF_GRID = [dict(n_estimators=300, max_depth=None),
-           dict(n_estimators=300, max_depth=12),
-           dict(n_estimators=600, max_depth=None)]
+# Use separate RF search spaces by task type.  Classification endpoints share
+# one small-sample regularized configuration, while regression and
+# biodistribution recover the original inner-CV grid.  Features and outer-fold
+# partitions are unchanged.
+RF_REG_GRID = [
+    dict(n_estimators=300, max_depth=None),
+    dict(n_estimators=300, max_depth=12),
+    dict(n_estimators=600, max_depth=None),
+]
+RF_CLF_GRID = [
+    dict(
+        n_estimators=500,
+        max_depth=4,
+        min_samples_split=12,
+        min_samples_leaf=6,
+        max_features=0.25,
+        bootstrap=True,
+        max_samples=0.70,
+    )
+]
 K_GRID = [3, 5, 10, 15]
 ALPHA_GRID = [0.3, 0.5, 0.7]
 
@@ -98,7 +138,14 @@ def rf_predict(tr, te, y, task_type, nc, params):
     if task_type == "reg":
         m = RandomForestRegressor(random_state=SEED, n_jobs=-1, **params); m.fit(Xtr, ytr); return m.predict(Xte)
     if task_type == "clf":
-        m = RandomForestClassifier(random_state=SEED, n_jobs=-1, **params); m.fit(Xtr, ytr); return m.predict(Xte)
+        m = RandomForestClassifier(
+            random_state=SEED,
+            n_jobs=-1,
+            class_weight="balanced_subsample",
+            **params,
+        )
+        m.fit(Xtr, ytr)
+        return m.predict(Xte)
     m = RandomForestRegressor(random_state=SEED, n_jobs=-1, **params); m.fit(Xtr, ytr)
     v = np.clip(m.predict(Xte), 0, None); s = v.sum(1, keepdims=True); return np.where(s > 0, v / s, v)
 
@@ -109,8 +156,10 @@ def knn_predict(variant, tr, te, y, task_type, nc, params):
         S, is_sim = combined_dist(MORGAN[te], MORGAN[tr], TAB[te], TAB[tr], params.get("alpha", 0.5)), False
     return knn_from_matrix(S, is_sim, y[tr], params["k"], task_type, nc)
 
-def grid_for(model):
-    if model == "rf": return [{"rf": g} for g in RF_GRID]
+def grid_for(model, task_type):
+    if model == "rf":
+        grid = RF_CLF_GRID if task_type == "clf" else RF_REG_GRID
+        return [{"rf": g} for g in grid]
     if model == "tanimoto_knn": return [{"k": k} for k in K_GRID]
     return [{"k": k, "alpha": a} for k in K_GRID for a in ALPHA_GRID]
 
@@ -119,19 +168,41 @@ def predict(model, tr, te, y, task_type, nc, params):
         else knn_predict(model, tr, te, y, task_type, nc, params)
 
 def select_params(model, tr, y, valid, task_type, nc, n_inner):
-    cand = grid_for(model)
+    cand = grid_for(model, task_type)
     if len(cand) == 1: return cand[0]
-    n_splits = max(2, min(n_inner, len(tr) // 2))
-    kf = KFold(n_splits=n_splits, shuffle=True, random_state=SEED)
+    base = tr[valid[tr]]
+    if len(base) < 4:
+        return cand[0]
+    if task_type == "clf":
+        class_counts = np.bincount(y[base].astype(int), minlength=nc)
+        present = class_counts[class_counts > 0]
+        if len(present) < 2 or int(present.min()) < 2:
+            return cand[0]
+        n_splits = min(n_inner, int(present.min()))
+    else:
+        n_splits = max(2, min(n_inner, len(base) // 2))
     best, best_s = cand[0], -1e18
     for p in cand:
         scs = []
-        for itr, iva in kf.split(tr):
-            a, b = tr[itr], tr[iva]
-            a, b = a[valid[a]], b[valid[b]]
+        # Recreate the deterministic split iterator for every candidate.
+        if task_type == "clf":
+            split_iter = StratifiedKFold(
+                n_splits=n_splits, shuffle=True, random_state=SEED
+            ).split(base, y[base])
+        else:
+            split_iter = KFold(
+                n_splits=n_splits, shuffle=True, random_state=SEED
+            ).split(base)
+        for itr, iva in split_iter:
+            a, b = base[itr], base[iva]
             if len(a) < 2 or len(b) < 1: continue
             try:
-                scs.append(score(task_type, y[b], predict(model, a, b, y, task_type, nc, p)))
+                scs.append(score(
+                    task_type,
+                    y[b],
+                    predict(model, a, b, y, task_type, nc, p),
+                    nc,
+                ))
             except Exception:
                 pass
         if scs and float(np.mean(scs)) > best_s:
@@ -181,10 +252,14 @@ def main():
                 trv, tev = tr[valid[tr]], te[valid[te]]
                 if len(trv) < 2 or len(tev) < 1: continue
                 params = select_params(model, tr, y, valid, ttype, nc, args.n_inner)
-                m = METRICS[ttype](y[tev], predict(model, trv, tev, y, ttype, nc, params))
+                pred = predict(model, trv, tev, y, ttype, nc, params)
+                if ttype == "clf":
+                    m = clf_metrics(y[tev], pred, nc)
+                else:
+                    m = METRICS[ttype](y[tev], pred)
                 tm[name] = m
                 for mk, mv in m.items():
-                    if mk == "n_samples": continue
+                    if mk == "n_samples" or not np.isscalar(mv): continue
                     agg.setdefault(name, {}).setdefault(mk, []).append(mv)
             fold_results.append({"fold": k, "test_metrics": tm})
             fdir = Path(args.out_root) / model / f"seed{args.seed}" / f"outer_fold_{k}"
@@ -203,4 +278,4 @@ def main():
         print(f"[{model}] saved -> {run_dir / 'summary.json'}")
 
 if __name__ == "__main__":
-    main()
\ No newline at end of file
+    main()