fix the answer parsing not aligned logic

chatkit/backend/app/answer_grid.py

@@ -15,7 +15,10 @@ import re
 from dataclasses import dataclass
 from typing import Optional
 
-_BLANK_MARKERS = {"", "-", "–", "—", "n/a", "none", "null"}
+_BLANK_MARKERS = {"", "n/a", "none", "null"}
+_CORRECT_MARKER = "="
+# Legacy dash markers are accepted as input and converted to CORRECT_MARKER
+_DASH_MARKERS = {"-", "–", "—"}
 _LETTER_RE = re.compile(r"^[A-Z]$")
 
 
@@ -48,12 +51,19 @@ class WrongAnswer:
 
 
 def normalize_letter(raw: Optional[str]) -> Optional[str]:
-    """Normalize a raw cell to a single uppercase letter or None."""
+    """Normalize a raw cell to a letter (A-Z), the correct-marker "=", or None.
+
+    - "=" or any legacy dash marker ("-", "–", "—") → "=" (answered correctly)
+    - Single letters with optional trailing "A) foo" → uppercase letter
+    - Empty/blank/unparseable → None
+    """
     if raw is None:
         return None
     s = str(raw).strip()
-    if s.casefold() in _BLANK_MARKERS:
+    if not s or s.casefold() in _BLANK_MARKERS:
         return None
+    if s == _CORRECT_MARKER or s in _DASH_MARKERS:
+        return _CORRECT_MARKER
     s = s.upper()
     # Tolerate "A)" or "A) foo" -> "A"
     if len(s) > 1 and s[0].isalpha() and s[1] in (")", ".", "、", " "):
@@ -147,7 +157,7 @@ def seed_from_parsed(
     - N is derived from questions_data if present, else from teacher_answers length,
       else from the longest student row.
     - Student rows are padded/truncated to N.
-    - Cells are normalized to letters or None.
+    - Cells are normalized to letters, the correct-marker "=", or None.
     """
     q_list = (questions_data or {}).get("questions") or []
     t_list = (teacher_answers_data or {}).get("answers") or []
@@ -175,7 +185,8 @@ def seed_from_parsed(
         if isinstance(qn, int) and 1 <= qn <= n:
             official[qn - 1] = normalize_letter(a.get("correct_answer"))
 
-    # Build student rows by question_number index
+    # Build student rows by question_number index. Legacy "-" markers are
+    # mapped to the canonical "=" correct-marker by normalize_letter.
     students: list[StudentRow] = []
     for i, s in enumerate(s_list):
         row: list[Optional[str]] = [None] * n
@@ -207,17 +218,20 @@ def seed_from_parsed(
 def diff_student(grid: AnswerGrid, student_index: int) -> list[WrongAnswer]:
     """Return the student's wrong answers ordered by question number.
 
-    Skips questions where the official key is blank (None).
+    - Skips questions where the official key is blank (None) or "=".
+    - Skips questions where the student's cell is "=" (answered correctly).
+    - Otherwise flags any cell that differs from the official letter.
     """
     if student_index < 0 or student_index >= len(grid.students):
         raise IndexError(f"student_index {student_index} out of range")
     row = grid.students[student_index].answers
     wrongs: list[WrongAnswer] = []
     for i, correct in enumerate(grid.official_answers):
-        if correct is None:
+        if correct is None or correct == _CORRECT_MARKER:
             continue
         student_ans = row[i] if i < len(row) else None
-        # Blank student answer is always wrong when a key exists
+        if student_ans == _CORRECT_MARKER:
+            continue
         if student_ans != correct:
             wrongs.append(
                 WrongAnswer(

chatkit/backend/app/file_processor.py

@@ -52,15 +52,15 @@ IMPORTANT - Images, Diagrams, Charts, and Figures:
 EXTRACTION_PROMPTS = {
     "questions": """You are an exam data extraction expert. Extract all exam questions from the provided content.
 
+This system supports SINGLE-CHOICE questions only (one correct letter: A, B, C, D, ...).
+
 Return a JSON object with this structure:
 {
   "questions": [
     {
       "number": 1,
       "text": "the full question text (including [圖片描述: ...] for any images/diagrams)",
-      "type": "multiple_choice" or "short_answer" or "essay" or "true_false",
-      "options": ["A) ...", "B) ...", "C) ...", "D) ..."],  // null if not multiple choice
-      "points": 10  // point value if visible, null otherwise
+      "options": ["A) ...", "B) ...", "C) ...", "D) ..."]
     }
   ]
 }
@@ -68,12 +68,14 @@ Return a JSON object with this structure:
 Rules:
 - Extract EVERY question you can see
 - Preserve the original language of the questions
-- If options are present, include them exactly as written
-- If point values are shown, include them
+- Always include the options exactly as written; every question must have options
+- Do NOT include fields other than number/text/options
 """ + IMAGE_DESCRIPTION_RULE + NO_FABRICATION_RULE + "- Return valid JSON only",
 
     "student_answers": """You are an exam data extraction expert. Extract all student answers from the provided content.
 
+This system supports SINGLE-CHOICE questions only.
+
 Return a JSON object with this structure:
 {
   "students": [
@@ -83,7 +85,7 @@ Return a JSON object with this structure:
       "answers": [
         {
           "question_number": 1,
-          "answer": "the student's answer text"
+          "answer": "A" | "B" | "C" | ... | "=" | null
         }
       ]
     }
@@ -92,20 +94,23 @@ Return a JSON object with this structure:
 
 Rules:
 - Extract answers for EVERY student visible
-- Preserve exact answer text
-- If a student left a question blank, set answer to null
+- Answer format (use exactly one of these):
+  - A single uppercase letter (A, B, C, D, ...) — the letter the student chose
+  - "=" — the student answered correctly (e.g., source uses "-" / "✓" / "correct" / matches the answer key)
+  - null — the student left the question blank
 - Use the student's name or ID as shown in the document
-- If a student's answer includes a drawing or diagram, describe it as text: [圖片描述: ...]
-""" + IMAGE_DESCRIPTION_RULE + NO_FABRICATION_RULE + "- Return valid JSON only",
+""" + NO_FABRICATION_RULE + "- Return valid JSON only",
 
     "teacher_answers": """You are an exam data extraction expert. Extract the correct/model answers (answer key) from the provided content.
 
+This system supports SINGLE-CHOICE questions only.
+
 Return a JSON object with this structure:
 {
   "answers": [
     {
       "question_number": 1,
-      "correct_answer": "the correct answer",
+      "correct_answer": "A" | "B" | "C" | ...,
       "explanation": "explanation if provided, otherwise null"
     }
   ]
@@ -113,10 +118,9 @@ Return a JSON object with this structure:
 
 Rules:
 - Extract the correct answer for EVERY question
+- correct_answer MUST be a single uppercase letter (A, B, C, D, ...)
 - Include explanations if they are provided
-- Preserve exact answer text
-- If the answer references or includes an image/diagram, describe it as text: [圖片描述: ...]
-""" + IMAGE_DESCRIPTION_RULE + NO_FABRICATION_RULE + "- Return valid JSON only",
+""" + NO_FABRICATION_RULE + "- Return valid JSON only",
 }
 
 

chatkit/backend/tests/test_answer_grid.py

@@ -30,10 +30,12 @@ from app.answer_grid import (
         (" b ", "B"),
         ("C)", "C"),
         ("D) foo", "D"),
+        ("=", "="),
+        (" = ", "="),
+        ("-", "="),
+        ("–", "="),
+        ("—", "="),
         ("", None),
-        ("-", None),
-        ("–", None),
-        ("—", None),
         (None, None),
         ("   ", None),
         ("N/A", None),
@@ -184,17 +186,41 @@ def test_seed_from_parsed_basic():
 
 
 @pytest.mark.unit
-def test_seed_from_parsed_handles_dash_as_blank():
+def test_seed_from_parsed_legacy_dash_becomes_correct_marker():
+    """Legacy '-' in a student answer normalizes to the canonical '=' marker."""
+    questions = {"questions": [{"number": 1, "text": "Q"}, {"number": 2, "text": "Q"}]}
+    students = {
+        "students": [
+            {
+                "name": "Alice",
+                "answers": [
+                    {"question_number": 1, "answer": "-"},
+                    {"question_number": 2, "answer": "D"},
+                ],
+            }
+        ]
+    }
+    teachers = {
+        "answers": [
+            {"question_number": 1, "correct_answer": "A"},
+            {"question_number": 2, "correct_answer": "B"},
+        ]
+    }
+    grid = seed_from_parsed(questions, students, teachers)
+    assert grid.students[0].answers == ("=", "D")
+
+
+@pytest.mark.unit
+def test_seed_from_parsed_preserves_explicit_equals():
     questions = {"questions": [{"number": 1, "text": "Q"}]}
     students = {
         "students": [
-            {"name": "Alice", "answers": [{"question_number": 1, "answer": "-"}]}
+            {"name": "Alice", "answers": [{"question_number": 1, "answer": "="}]}
         ]
     }
     teachers = {"answers": [{"question_number": 1, "correct_answer": "A"}]}
     grid = seed_from_parsed(questions, students, teachers)
-    # Dash normalizes to None — diff treats None != "A" as wrong
-    assert grid.students[0].answers == (None,)
+    assert grid.students[0].answers == ("=",)
 
 
 @pytest.mark.unit
@@ -263,6 +289,39 @@ def test_diff_student_blank_student_answer_is_wrong():
     assert wrongs[0].student_answer is None
 
 
+@pytest.mark.unit
+def test_diff_student_equals_marker_is_correct():
+    """A '=' cell means 'student answered correctly' and must never appear in wrongs."""
+    grid = AnswerGrid(
+        total_questions=3,
+        official_answers=("A", "B", "C"),
+        students=(StudentRow(name="Alice", answers=("=", "X", "=")),),
+        questions=tuple(
+            QuestionMeta(number=i + 1, text=f"Q{i+1}", options=()) for i in range(3)
+        ),
+    )
+    wrongs = diff_student(grid, 0)
+    assert len(wrongs) == 1
+    assert wrongs[0].question.number == 2
+    assert wrongs[0].student_answer == "X"
+
+
+@pytest.mark.unit
+def test_diff_student_skips_equals_in_official():
+    grid = AnswerGrid(
+        total_questions=2,
+        official_answers=("A", "="),  # Q2 has no authoritative key
+        students=(StudentRow(name="Alice", answers=("X", "Y")),),
+        questions=(
+            QuestionMeta(number=1, text="Q1", options=()),
+            QuestionMeta(number=2, text="Q2", options=()),
+        ),
+    )
+    wrongs = diff_student(grid, 0)
+    assert len(wrongs) == 1
+    assert wrongs[0].question.number == 1
+
+
 @pytest.mark.unit
 def test_diff_student_skips_blank_official_key():
     grid = AnswerGrid(

chatkit/frontend/src/components/step1/FileUploadPanel.tsx

@@ -274,7 +274,7 @@ function ZoneSummary({ type, data }: { type: string; data: Record<string, unknow
 
 function QuestionsTable({ data }: { data?: Record<string, unknown> }) {
   if (!data) return null;
-  const questions = (data as { questions?: Array<{ number: number; text: string; type?: string; options?: string[] | null; points?: number | null }> }).questions;
+  const questions = (data as { questions?: Array<{ number: number; text: string; options?: string[] | null }> }).questions;
   if (!questions?.length) return null;
 
   return (
@@ -288,8 +288,6 @@ function QuestionsTable({ data }: { data?: Record<string, unknown> }) {
             <tr className="border-b border-[var(--color-border)]">
               <th className="text-left py-2 px-3 text-[var(--color-text-muted)] w-12">#</th>
               <th className="text-left py-2 px-3 text-[var(--color-text-muted)]">題目</th>
-              <th className="text-left py-2 px-3 text-[var(--color-text-muted)] w-28">類型</th>
-              <th className="text-left py-2 px-3 text-[var(--color-text-muted)] w-16">分數</th>
             </tr>
           </thead>
           <tbody>
@@ -304,10 +302,6 @@ function QuestionsTable({ data }: { data?: Record<string, unknown> }) {
                     </div>
                   )}
                 </td>
-                <td className="py-2 px-3">
-                  {q.type && <span className="badge badge-success text-xs">{q.type}</span>}
-                </td>
-                <td className="py-2 px-3 text-[var(--color-text-muted)]">{q.points ?? "-"}</td>
               </tr>
             ))}
           </tbody>

chatkit/frontend/src/components/step2/AnswerGridEditor.tsx

@@ -29,11 +29,16 @@ interface AnswerGridEditorProps {
   onConfirmedChange: (grid: AnswerGrid | null) => void;
 }
 
+const CORRECT_MARKER = "=";
+const DASH_MARKERS = new Set(["-", "–", "—"]);
+
 function normalizeCell(raw: string): string | null {
-  const trimmed = raw.trim().toUpperCase();
-  if (!trimmed || trimmed === "-") return null;
-  // Take first letter only, allow A-Z
-  const first = trimmed.charAt(0);
+  const trimmed = raw.trim();
+  if (!trimmed) return null;
+  if (trimmed === CORRECT_MARKER || DASH_MARKERS.has(trimmed)) {
+    return CORRECT_MARKER;
+  }
+  const first = trimmed.charAt(0).toUpperCase();
   return /^[A-Z]$/.test(first) ? first : null;
 }
 
@@ -179,6 +184,7 @@ export function AnswerGridEditor({
           </h3>
           <p className="text-xs text-[var(--color-text-muted)] mt-1">
             請核對學生作答與標準答案，確認後才能生成報告。支援單選題（A-Z）。
+            學生欄輸入 <code className="px-1 bg-[var(--color-background)] rounded">=</code> 代表該題答對。
           </p>
         </div>
         <div className="flex items-center gap-2">
@@ -217,20 +223,28 @@ export function AnswerGridEditor({
                 <td className="py-1 px-3 border-b border-[var(--color-border)]/50 sticky left-0 bg-[var(--color-background)] z-10 font-medium text-[var(--color-text)]">
                   {s.name}
                 </td>
-                {qNums.map((n, qi) => (
-                  <td
-                    key={n}
-                    className="border-b border-[var(--color-border)]/50 p-0.5"
-                  >
-                    <input
-                      type="text"
-                      maxLength={2}
-                      defaultValue={s.answers[qi] ?? ""}
-                      onBlur={(e) => updateStudentCell(si, qi, e.target.value)}
-                      className="w-full text-center py-1 bg-transparent text-[var(--color-text)] border border-transparent hover:border-[var(--color-border)] focus:border-[var(--color-primary)] focus:outline-none rounded"
-                    />
-                  </td>
-                ))}
+                {qNums.map((n, qi) => {
+                  const cell = s.answers[qi];
+                  const isCorrect = cell === CORRECT_MARKER;
+                  return (
+                    <td
+                      key={n}
+                      className="border-b border-[var(--color-border)]/50 p-0.5"
+                    >
+                      <input
+                        type="text"
+                        maxLength={2}
+                        defaultValue={cell ?? ""}
+                        onBlur={(e) => updateStudentCell(si, qi, e.target.value)}
+                        className={`w-full text-center py-1 bg-transparent border border-transparent hover:border-[var(--color-border)] focus:border-[var(--color-primary)] focus:outline-none rounded ${
+                          isCorrect
+                            ? "text-[var(--color-success)] font-semibold"
+                            : "text-[var(--color-text)]"
+                        }`}
+                      />
+                    </td>
+                  );
+                })}
               </tr>
             ))}