Fix: Treat '-' and similar markers as correct answers in student data

Student answer data often uses '-', 'v', '✓' etc. to mark correct answers.
filter_student_wrong_answers now skips these markers and only includes
actual wrong answer text. Works with or without teacher answers uploaded.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

chatkit/backend/app/report_generator.py

@@ -47,15 +47,28 @@ def extract_html(content: str) -> str:
 # Per-student report generation (new flow)
 # ---------------------------------------------------------------------------
 
+# Markers that mean "correct / no wrong answer" in student answer data
+_CORRECT_MARKERS = {"-", "—", "–", "v", "V", "✓", "✔", "○", "O", "o", "正確"}
+
+
+def _is_correct_marker(answer: str) -> bool:
+    """Check if a student answer value means 'correct' (not a real wrong answer)."""
+    return answer.strip() in _CORRECT_MARKERS
+
+
 def filter_student_wrong_answers(
     student: dict,
     questions_data: dict,
     teacher_answers_data: dict,
 ) -> tuple[list[dict], list[dict], list[dict], bool]:
-    """Compare a student's answers with teacher answers, return wrong items.
+    """Find a student's wrong answers.
+
+    Detection strategy (in priority order):
+    1. If answer is a correct-marker ("-", "v", "✓", etc.) → skip (correct)
+    2. If teacher answers exist → compare against correct_answer
+    3. If no teacher answers and not a marker → treat as wrong
 
     Returns (wrong_questions, wrong_student_answers, relevant_teacher_answers, has_teacher_answers).
-    When teacher answers are missing, returns ALL student answers & questions instead.
     """
     teacher_map: dict[int, dict] = {}
     for ta in teacher_answers_data.get("answers", []):
@@ -69,11 +82,7 @@ def filter_student_wrong_answers(
         if num is not None:
             question_map[int(num)] = q
 
-    # No teacher answers → return ALL student answers for LLM to analyze
-    if not teacher_map:
-        all_questions = questions_data.get("questions", [])
-        all_answers = student.get("answers", [])
-        return all_questions, all_answers, [], False
+    has_teacher = bool(teacher_map)
 
     wrong_questions: list[dict] = []
     wrong_answers: list[dict] = []
@@ -84,21 +93,32 @@ def filter_student_wrong_answers(
         if qnum is None:
             continue
         qnum = int(qnum)
-        ta = teacher_map.get(qnum)
-        if ta is None:
+
+        student_ans = str(sa.get("answer") or "").strip()
+
+        # Skip blank answers
+        if not student_ans:
             continue
 
-        student_ans = str(sa.get("answer", "")).strip()
-        correct_ans = str(ta.get("correct_answer", "")).strip()
+        # "-" and similar markers mean correct → skip
+        if _is_correct_marker(student_ans):
+            continue
 
-        if student_ans.lower() != correct_ans.lower():
-            wrong_answers.append(sa)
+        # If we have teacher answers, double-check against correct answer
+        ta = teacher_map.get(qnum)
+        if ta:
+            correct_ans = str(ta.get("correct_answer", "")).strip()
+            if student_ans.lower() == correct_ans.lower():
+                continue  # actually correct
             relevant_teacher.append(ta)
-            q = question_map.get(qnum)
-            if q:
-                wrong_questions.append(q)
 
-    return wrong_questions, wrong_answers, relevant_teacher, True
+        # This is a wrong answer
+        wrong_answers.append(sa)
+        q = question_map.get(qnum)
+        if q:
+            wrong_questions.append(q)
+
+    return wrong_questions, wrong_answers, relevant_teacher, has_teacher
 
 
 STUDENT_REPORT_SYSTEM_PROMPT = """You are ClassLens, an AI teaching assistant that creates a focused, personalized exam analysis report for a single student.