Verbal flashcards word length dependent permission

zeeguu/api/test/test_verbal_flashcards.py

@@ -304,16 +304,17 @@ def test_score_word_match_accepts_common_danish_asr_variants():
 
 
 @pytest.mark.parametrize(
-    "user_word, expected_word",
+    "user_word, expected_word, expected_allowed_distance",
     [
-        ("hat", "kat"),
-        ("hond", "hund"),
-        ("pange", "penge"),
+        ("hat", "kat", 1),
+        ("hond", "hund", 1),
+        ("pange", "penge", 2),
     ],
 )
-def test_score_word_match_accepts_one_optimal_string_alignment_edit(
+def test_score_word_match_accepts_words_within_length_based_edit_budget(
     user_word,
     expected_word,
+    expected_allowed_distance,
 ):
     from zeeguu.core.verbal_flashcards.fuzzy_match import score_word_match
 
@@ -322,33 +323,37 @@ def test_score_word_match_accepts_one_optimal_string_alignment_edit(
     assert result["isMatch"] is True
     assert result["matchType"] == "fuzzy"
     assert result["optimalStringAlignmentDistance"] == 1
-    assert result["allowedOptimalStringAlignmentDistance"] == 1
+    assert result["allowedOptimalStringAlignmentDistance"] == expected_allowed_distance
     assert result["jaroWinkler"] > 0
 
 
 @pytest.mark.parametrize(
-    "user_word, expected_word",
+    "user_word, expected_word, expected_allowed_distance",
     [
-        ("hot", "kat"),
-        ("hd", "hund"),
-        ("pen", "penge"),
+        ("hot", "kat", 1),
+        ("zzzz", "hund", 1),
+        ("xxxxx", "penge", 2),
     ],
 )
-def test_score_word_match_rejects_multiple_optimal_string_alignment_edits(
+def test_score_word_match_rejects_words_outside_length_based_edit_budget(
     user_word,
     expected_word,
+    expected_allowed_distance,
 ):
     from zeeguu.core.verbal_flashcards.fuzzy_match import score_word_match
 
     result = score_word_match(user_word, expected_word, language_code="da")
 
     assert result["isMatch"] is False
     assert result["matchType"] == "close"
-    assert result["optimalStringAlignmentDistance"] > 1
-    assert result["allowedOptimalStringAlignmentDistance"] == 1
+    assert (
+        result["optimalStringAlignmentDistance"]
+        > result["allowedOptimalStringAlignmentDistance"]
+    )
+    assert result["allowedOptimalStringAlignmentDistance"] == expected_allowed_distance
 
 
-def test_score_word_match_requires_exact_match_for_two_letter_words():
+def test_score_word_match_allows_one_edit_for_two_letter_words():
     from zeeguu.core.verbal_flashcards.fuzzy_match import score_word_match
 
     result = score_word_match("og", "ok", language_code="da")

zeeguu/core/verbal_flashcards/fuzzy_match.py

@@ -152,14 +152,22 @@ def allowed_optimal_string_alignment_distance(expected_word, language_code=None)
     Return the maximum edit distance accepted for a spoken flashcard answer.
 
     Acceptance is based on edit distance, not a blended similarity score:
-    after language-specific normalization, words of length >= 3 may differ by
-    one optimal string alignment edit. Jaro-Winkler is still returned as a
-    diagnostic signal for debugging and future analysis, but it does not decide
-    correctness.
+    after language-specific normalization, longer words get a larger edit
+    budget because ASR approximations often drift more on longer Danish words.
+    Jaro-Winkler is still returned as a diagnostic signal for debugging and
+    future analysis, but it does not decide correctness.
     """
     normalizer = normalizer_for(language_code)
     normalized_length = len(normalizer.canonical_form(expected_word))
-    return 0 if normalized_length <= 2 else 1
+    if normalized_length <= 2:
+        return 0
+    if normalized_length <= 4:
+        return 1
+    if normalized_length <= 6:
+        return 2
+    if normalized_length <= 9:
+        return 3
+    return 4
 
 
 def fuzzy_match_threshold(expected_word, language_code=None):