Update

metrics.py

@@ -21,7 +21,7 @@ def _parse_answer(text: str) -> List[List[str]]:
     alternatives = re.split(r';', text)
     result = [ ]
     for alternative in alternatives:
-        groups = re.split(r'-->|,', alternative)
+        groups = re.split(r'–?-?-?>|,', alternative)
         result.append([" ".join(re.findall(r'\b\w+\b', group)) for group in groups])
     return result
 
@@ -42,7 +42,7 @@ def _check_answer(completion: str, answer: str) -> bool:
     """
     completion = _answer_without_thoughts(completion).lower()
     completion  = re.sub(r'[^\w\s]', ' ', completion) # this replaces punctuations with space, aligning with the _parse_answer function's ' '.join
-    completion = re.sub(r'[\u2000-\u200F\u202F\u205F\u3000]', ' ', completion) # replaces unicode spaces with space
+    completion = re.sub(r'\s+', ' ', completion) # normalize consecutive (Unicode) spaces to finish aligning with _parse_answer
     alternative_answers = _parse_answer(answer)
     for answer_phrases in alternative_answers:
         # if all(phrase in completion for phrase in answer_phrases):

puzzles_cleaned.csv

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a008da9d40adfebfa2e0fafa504aba03125c9064efd77b45dbfcb68655eb878c
-size 1116075
+oid sha256:0a97676e6cfc842a0fe5cd0999a16c6523e71ef0692b125cd426cff963eabc1a
+size 1117498

results.duckdb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7d3a9c4a997e0b6741249ba973be3c145b6f660381633d35e5eaa94353ea30f
-size 39333888
+oid sha256:569fd5c2d797ed702e5d2551bc6a9d4d54df78481ee016bc9ee07767361525b8
+size 137637888