Add format-tax macro-F1 vs accuracy analysis

analysis/format_tax_macro_f1.json

@@ -0,0 +1,187 @@
+{
+  "metric": "viewpoint-family macro-F1 (%) and binary accuracy (%)",
+  "by_model": {
+    "Opus 4.5": {
+      "raw_prose": {
+        "acc": 97.0,
+        "macro_f1": 91.2,
+        "n": 166
+      },
+      "raw_text": {
+        "acc": 85.5,
+        "macro_f1": 85.1,
+        "n": 166
+      },
+      "shuffled": {
+        "acc": 78.9,
+        "macro_f1": 77.1,
+        "n": 166
+      },
+      "dsl": {
+        "acc": 68.7,
+        "macro_f1": 62.6,
+        "n": 166
+      },
+      "code": {
+        "acc": 75.3,
+        "macro_f1": 71.2,
+        "n": 166
+      },
+      "label_list": {
+        "acc": 69.9,
+        "macro_f1": 67.3,
+        "n": 166
+      }
+    },
+    "GPT-4o": {
+      "raw_prose": {
+        "acc": 87.3,
+        "macro_f1": 83.8,
+        "n": 166
+      },
+      "raw_text": {
+        "acc": 78.9,
+        "macro_f1": 81.8,
+        "n": 166
+      },
+      "shuffled": {
+        "acc": 75.9,
+        "macro_f1": 72.9,
+        "n": 166
+      },
+      "dsl": {
+        "acc": 67.5,
+        "macro_f1": 62.0,
+        "n": 166
+      },
+      "code": {
+        "acc": 61.4,
+        "macro_f1": 52.8,
+        "n": 166
+      },
+      "label_list": {
+        "acc": 75.9,
+        "macro_f1": 68.2,
+        "n": 166
+      }
+    },
+    "Gemini 2.5 Pro": {
+      "raw_prose": {
+        "acc": 98.6,
+        "macro_f1": 93.9,
+        "n": 295
+      },
+      "raw_text": {
+        "acc": 93.9,
+        "macro_f1": 91.2,
+        "n": 295
+      },
+      "shuffled": {
+        "acc": 86.1,
+        "macro_f1": 81.4,
+        "n": 295
+      },
+      "dsl": {
+        "acc": 78.3,
+        "macro_f1": 71.2,
+        "n": 295
+      },
+      "code": {
+        "acc": 81.7,
+        "macro_f1": 71.0,
+        "n": 295
+      },
+      "label_list": {
+        "acc": 82.4,
+        "macro_f1": 71.3,
+        "n": 295
+      }
+    },
+    "Kimi K2": {
+      "raw_prose": {
+        "acc": 88.0,
+        "macro_f1": 86.7,
+        "n": 166
+      },
+      "raw_text": {
+        "acc": 73.5,
+        "macro_f1": 75.8,
+        "n": 166
+      },
+      "shuffled": {
+        "acc": 66.9,
+        "macro_f1": 58.0,
+        "n": 166
+      },
+      "dsl": {
+        "acc": 69.9,
+        "macro_f1": 57.6,
+        "n": 166
+      },
+      "code": {
+        "acc": 61.4,
+        "macro_f1": 59.9,
+        "n": 166
+      },
+      "label_list": {
+        "acc": 69.9,
+        "macro_f1": 57.5,
+        "n": 166
+      }
+    },
+    "Llama 3.3 70B": {
+      "raw_prose": {
+        "acc": 87.3,
+        "macro_f1": 82.5,
+        "n": 166
+      },
+      "raw_text": {
+        "acc": 77.1,
+        "macro_f1": 77.1,
+        "n": 166
+      },
+      "shuffled": {
+        "acc": 65.7,
+        "macro_f1": 59.6,
+        "n": 166
+      },
+      "dsl": {
+        "acc": 60.8,
+        "macro_f1": 55.7,
+        "n": 166
+      },
+      "code": {
+        "acc": 59.0,
+        "macro_f1": 57.8,
+        "n": 166
+      },
+      "label_list": {
+        "acc": 57.2,
+        "macro_f1": 50.1,
+        "n": 166
+      }
+    }
+  },
+  "format_tax_gap": {
+    "Opus 4.5": {
+      "acc_gap": 28.3,
+      "macro_f1_gap": 28.6
+    },
+    "GPT-4o": {
+      "acc_gap": 25.9,
+      "macro_f1_gap": 31.0
+    },
+    "Gemini 2.5 Pro": {
+      "acc_gap": 20.3,
+      "macro_f1_gap": 22.9
+    },
+    "Kimi K2": {
+      "acc_gap": 26.6,
+      "macro_f1_gap": 29.2
+    },
+    "Llama 3.3 70B": {
+      "acc_gap": 30.1,
+      "macro_f1_gap": 32.4
+    }
+  }
+}