diff --git a/PR.md b/PR.md
new file mode 100644
index 0000000..0f53fd8
--- /dev/null
+++ b/PR.md
@@ -0,0 +1,83 @@
+# Fix `truncate` dropping a character when a string separator is absent
+
+## Problem
+
+`pydash.truncate(text, length, omission, separator)` is documented to truncate
+`text` at the last occurrence of `separator`. When `separator` is a **string
+that does not occur** in the truncated text, `truncate` incorrectly removes an
+extra trailing character instead of falling back to a plain cut.
+
+```python
+>>> import pydash
+>>> pydash.truncate("hello world", 10, "...", "xyz")
+'hello ...'      # expected 'hello w...'
+>>> pydash.truncate("hello world", 8, "...", "q")
+'hell...'        # expected 'hello...'
+```
+
+This is inconsistent with the equivalent lodash behaviour (the API pydash
+mirrors), where a separator that is not found leaves the plain truncation
+untouched:
+
+```js
+_.truncate('hello world', {length: 10, separator: 'xyz'}) // 'hello w...'
+_.truncate('hello world', {length: 8,  separator: 'q'})   // 'hello...'
+```
+
+It is also internally inconsistent: the regular-expression `separator` branch
+already handles "no match" correctly (it leaves `trunc_len` at the full
+truncated length), while the string branch does not.
+
+## Root cause
+
+`src/pydash/strings.py`, in `truncate`:
+
+```python
+if pyd.is_string(separator):
+    trunc_len = text.rfind(separator)
+```
+
+`str.rfind` returns `-1` when the separator is not present. That `-1` is then
+used as a slice end (`text[:trunc_len]` == `text[:-1]`), which silently drops
+the last character of the already-truncated text before appending `omission`.
+
+## Fix
+
+Only move the truncation point when the separator is actually found:
+
+```python
+if pyd.is_string(separator):
+    separator_index = text.rfind(separator)
+    if separator_index >= 0:
+        trunc_len = separator_index
+```
+
+This mirrors the existing regex branch, which only updates `trunc_len` when a
+match exists.
+
+## Test
+
+Added two cases to the existing `test_truncate` parametrization in
+`tests/test_strings.py` covering a string separator that does not occur in the
+truncated text:
+
+```python
+(("hello world", 10, "...", "xyz"), "hello w..."),
+(("hello world", 8, "...", "q"), "hello..."),
+```
+
+## Gate
+
+- `python -m pytest tests/test_strings.py -k test_truncate` -> 13 passed
+- `python -m pytest tests/ --ignore=tests/pytest_mypy_testing` -> 1917 passed
+
+Counterfactual (fix reverted, test kept): the new cases fail with
+`AssertionError: assert 'hello ...' == 'hello w...'` and
+`AssertionError: assert 'hell...' == 'hello...'`.
+
+## Flags for the account owner
+
+The `tests/pytest_mypy_testing/` suite fails locally with `NameError` (the
+`reveal_type` typing plugin is not resolvable in a bare `pip install -e .`
+environment). These failures are pre-existing and unrelated to this change,
+which touches only `truncate`.
diff --git a/src/pydash/strings.py b/src/pydash/strings.py
index 4bb779c..f442b7f 100644
--- a/src/pydash/strings.py
+++ b/src/pydash/strings.py
@@ -2064,7 +2064,9 @@ def truncate(
     trunc_len = len(text)
 
     if pyd.is_string(separator):
-        trunc_len = text.rfind(separator)
+        separator_index = text.rfind(separator)
+        if separator_index >= 0:
+            trunc_len = separator_index
     elif pyd.is_reg_exp(separator):
         last = None
         for match in separator.finditer(text):
diff --git a/tests/test_strings.py b/tests/test_strings.py
index 58680f2..6460686 100644
--- a/tests/test_strings.py
+++ b/tests/test_strings.py
@@ -1079,6 +1079,10 @@ def test_trim_end(case, expected):
             ("hi-diddly-ho there, neighborino", 24, "...", re.compile(",? +")),
             "hi-diddly-ho there...",
         ),
+        # A string separator that does not occur in the truncated text must not
+        # trim any additional characters; truncation falls back to the plain cut.
+        (("hello world", 10, "...", "xyz"), "hello w..."),
+        (("hello world", 8, "...", "q"), "hello..."),
         (("hi-diddly-ho there, neighborino", 30, " [...]"), "hi-diddly-ho there, neig [...]"),
         (("123456789", 9), "123456789"),
         (("123456789", 8), "12345..."),