diff --git a/stratum/tests/adapters/test_string_encoder.py b/stratum/tests/adapters/test_string_encoder.py
index 1c963a30..614e6fdb 100644
--- a/stratum/tests/adapters/test_string_encoder.py
+++ b/stratum/tests/adapters/test_string_encoder.py
@@ -5,7 +5,13 @@
 
 from stratum import StringEncoder
 from stratum import set_config
-from stratum.adapters.string_encoder import RustyStringEncoder, _clean_strings
+from stratum.adapters.string_encoder import (
+    RustyStringEncoder,
+    _clean_strings,
+    _rust_supported_subset,
+    _prep_strings,
+    _prep_strings_transform,
+)
 
 set_config(rust_backend=True, debug_timing=True, num_threads=8)
 
@@ -18,19 +24,104 @@ def capture_std_out(capfd):
     return combined_output
 
 @pytest.mark.parametrize("analyzer", ["char", "char_wb"])
-def test_string_encoder_result(analyzer, capfd):
+@pytest.mark.parametrize("vectorizer", ["hashing", "tfidf"])
+def test_string_encoder_result(analyzer, vectorizer, capfd):
     s = pd.Series(["foo", "bar", None, "lorem ipsum dolor"]) # nulls handled upstream
 
     # StringEncoder should point to our subclass
     assert StringEncoder is RustyStringEncoder
 
-    enc = StringEncoder(vectorizer='hashing', analyzer=analyzer, ngram_range=(3,5), n_components=2)
+    enc = StringEncoder(vectorizer=vectorizer, analyzer=analyzer, ngram_range=(3,5), n_components=2)
     Z = enc.fit_transform(s)
     assert Z.shape[0] == len(s)
 
+    # Round-trip through transform to cover the transform code path
+    s2 = pd.Series(["foo", "baz", "lorem"])
+    Z2 = enc.transform(s2)
+    assert Z2.shape[0] == len(s2)
+    assert Z2.shape[1] == enc.n_components_
+
     # Assert if rust timing appeared (verifies that rust code is executed)
     assert "[rust]" in capture_std_out(capfd)
 
+
+def test_string_encoder_padding():
+    """n_components larger than achievable rank triggers the padding branch."""
+    s = pd.Series(["a", "b"])
+    enc = StringEncoder(vectorizer="hashing", analyzer="char",
+                        ngram_range=(3, 5), n_components=50)
+    Z = enc.fit_transform(s)
+    assert Z.shape == (2, 50)
+    Z2 = enc.transform(pd.Series(["a", "c"]))
+    assert Z2.shape == (2, 50)
+
+
+def test_string_encoder_fallback_when_rust_disabled():
+    """Disabling rust_backend routes through sklearn fallback path."""
+    set_config(rust_backend=False)
+    try:
+        s = pd.Series(["foo", "bar", "baz", "lorem ipsum"])
+        enc = StringEncoder(vectorizer="hashing", analyzer="char",
+                            ngram_range=(3, 5), n_components=2)
+        Z = enc.fit_transform(s)
+        assert Z.shape[0] == len(s)
+        Z2 = enc.transform(pd.Series(["foo", "bar"]))
+        assert Z2.shape[0] == 2
+    finally:
+        set_config(rust_backend=True)
+
+
+def test_transform_without_fit_falls_back():
+    """Calling transform before fit_transform should take the sklearn fallback."""
+    enc = StringEncoder(vectorizer="hashing", analyzer="char",
+                        ngram_range=(3, 5), n_components=2)
+    # Force sklearn fit so internal sklearn state exists, but _rust_state_ stays None
+    set_config(rust_backend=False)
+    try:
+        enc.fit(pd.Series(["foo", "bar", "baz"]))
+    finally:
+        set_config(rust_backend=True)
+    # _rust_state_ is None → transform should fall back
+    Z = enc.transform(pd.Series(["foo", "bar"]))
+    assert Z.shape[0] == 2
+
+
+class _Dummy:
+    def __init__(self, **kw):
+        for k, v in kw.items():
+            setattr(self, k, v)
+
+
+def test_rust_supported_subset_rejections():
+    ok, _ = _rust_supported_subset(_Dummy(vectorizer="hashing", stop_words=None,
+                                         analyzer="char", ngram_range=(3, 5)))
+    assert ok
+
+    ok, msg = _rust_supported_subset(_Dummy(vectorizer="count"))
+    assert not ok and "vectorizer" in msg
+
+    ok, msg = _rust_supported_subset(_Dummy(vectorizer="tfidf", stop_words="english"))
+    assert not ok and "stop_words" in msg
+
+    ok, msg = _rust_supported_subset(_Dummy(vectorizer="tfidf", stop_words=None,
+                                            analyzer="word"))
+    assert not ok and "analyzer" in msg
+
+    ok, msg = _rust_supported_subset(_Dummy(vectorizer="tfidf", stop_words=None,
+                                            analyzer="char", ngram_range=(5, 3)))
+    assert not ok and "ngram_range" in msg
+
+
+def test_prep_strings_roundtrip():
+    s = pd.Series(["foo", None, "bar"])
+    out = _prep_strings(s)
+    assert len(out) == 3
+    assert all(isinstance(v, str) for v in out)
+
+    out2 = _prep_strings_transform(s)
+    assert len(out2) == 3
+    assert all(isinstance(v, str) for v in out2)
+
 def test_clean_strings_with_nan():
     """Test _clean_strings with NaN values."""
     x_list = ["foo", None, np.nan, float('nan'), "bar", 123]