feat(graders): add letter grade, summary and full metrics dict to grade_episode()

graders/programmatic_grader.py

@@ -7,16 +7,35 @@ from graders.easy_grader import grade_easy
 from graders.medium_grader import grade_medium
 from graders.hard_grader import grade_hard
 
+def get_letter_grade(score: float) -> str:
+    if score >= 0.90: return "A+"
+    if score >= 0.80: return "A"
+    if score >= 0.70: return "B"
+    if score >= 0.60: return "C"
+    if score >= 0.50: return "D"
+    return "F"
+
 def grade_episode(history: list[dict]) -> dict[str, Any]:
     if not history:
-        return {"error": "Empty history"}
+        return {"error": "No history provided", "score": 0.0, "grade": "F"}
     
-    easy_score = grade_easy(history)
+    easy_score   = grade_easy(history)
     medium_score = grade_medium(history)
-    hard_score = grade_hard(history)
+    hard_score   = grade_hard(history)
+    
+    final_score = hard_score
+    grade       = get_letter_grade(final_score)
     
     return {
-        "easy_score": easy_score,
-        "medium_score": medium_score,
-        "hard_score": hard_score
+        "overall_score": round(final_score, 4),
+        "letter_grade":  grade,
+        "summary":       f"Episode completed with {len(history)} turns. Final score: {final_score}.",
+        "metrics": {
+            "easy":   easy_score,
+            "medium": medium_score,
+            "hard":   hard_score,
+            "turns":  len(history),
+            "unique_strategies": len(set(h.get("strategy_type") for h in history)),
+            "unique_categories": len(set(h.get("target_category") for h in history)),
+        }
     }