fix(v1): prefer final judge choices over draft boxes

tests/v1/test_scoring.py

@@ -113,7 +113,39 @@ def test_parse_pytest_outcomes_strips_xfail_xpass_reasons() -> None:
     }
 
 
-def test_parse_judge_choice_uses_first_choice_after_verdict_marker() -> None:
-    response = "Final Judgment: B because it is a better answer"
-
-    assert vf.parse_judge_choice(response, choices=("A", "B")) == "B"
+def test_parse_judge_choice_prefers_final_marker_then_boxed() -> None:
+    assert (
+        vf.parse_judge_choice(
+            "Draft: \\boxed{A}\nFinal Judgment: B", choices=("A", "B")
+        )
+        == "B"
+    )
+    assert (
+        vf.parse_judge_choice(
+            "Draft: \\boxed{A}\nFinal Judgment:\nReasoning mentions B",
+            choices=("A", "B"),
+        )
+        == "A"
+    )
+    assert (
+        vf.parse_judge_choice(
+            "Draft: \\boxed{B}\nFinal Judgment: This is a \\boxed{B}",
+            choices=("A", "B"),
+        )
+        == "B"
+    )
+    assert (
+        vf.parse_judge_choice(
+            "Final Judgment: A better choice is \\boxed{B}", choices=("A", "B")
+        )
+        == "A"
+    )
+    assert (
+        vf.parse_judge_choice(
+            "Final Judgment: \\boxed{A}\nFinal Judgment: B", choices=("A", "B")
+        )
+        == "B"
+    )
+    assert (
+        vf.parse_judge_choice("Draft: \\boxed{A}\nAnswer: B", choices=("A", "B")) == "A"
+    )

verifiers/v1/utils/score.py

@@ -55,12 +55,23 @@ def parse_judge_choice(
         return None
 
     text = content.rsplit("</think>", 1)[-1].strip()
-    text = extract_boxed_answer(text, strict=True).strip() or text
-
     text_upper = text.upper()
     choices_by_upper = {choice.upper(): choice for choice in choices}
     allowed = "|".join(re.escape(choice) for choice in choices_by_upper)
     choice_re = rf"(?<!\w)({allowed})(?!\w)"
+    horizontal_space = r"[^\S\n]"
+    final_choices = re.findall(
+        rf"^{horizontal_space}*FINAL{horizontal_space}+"
+        rf"(?:JUDGMENT|ANSWER|VERDICT){horizontal_space}*"
+        rf"(?:IS{horizontal_space}*)?[:\-]?{horizontal_space}*{choice_re}",
+        text_upper,
+        re.MULTILINE,
+    )
+    if final_choices:
+        return choices_by_upper.get(final_choices[-1])
+
+    text = extract_boxed_answer(text, strict=True).strip() or text
+    text_upper = text.upper()
     verdict_re = (
         r"(?:^|\n)\s*(?:FINAL\s+JUDGMENT|FINAL\s+ANSWER|FINAL\s+VERDICT|"
         r"JUDGMENT|VERDICT|ANSWER)\s*(?:IS\s*)?[:\-]?\s*"