FIX: sample within word punctuation insertion from every position

frontend/src/components/Scenarios/ScenarioDetail.test.tsx

@@ -2023,7 +2023,8 @@ describe('ScenarioDetail', () => {
   })
 
   it.each(['user', 'unknown'])('warns before replacing %s data and confirms through Run preview', async (origin) => {
-    const user = userEvent.setup()
+    jest.useFakeTimers()
+    const user = userEvent.setup({ advanceTimers: jest.advanceTimersByTime })
     jest.mocked(datasetsApi.listDatasets).mockResolvedValueOnce({ items: [{
       name: 'harmbench', selection_key: 'dataset:named:harmbench', loaded: true, provider_available: true,
       logical_examples: 8, origins: [origin], modalities: ['text'], harm_categories: [], has_unlabeled_harm_categories: false,
@@ -2035,7 +2036,9 @@ describe('ScenarioDetail', () => {
       },
     })
     renderDetail('/scanner/foundry.red_team_agent')
+    await flushRenderedPromises()
     await user.click(await screen.findByRole('radio', { name: 'Refresh' }))
+    await advanceTimers(300)
     await user.click(screen.getByRole('button', { name: 'Launch scan' }))
     const preview = await screen.findByRole('dialog', { name: 'Run preview' })
     await waitFor(() => expect(within(preview).getByRole('button', { name: 'Launch scan' })).toBeEnabled())
@@ -2044,7 +2047,7 @@ describe('ScenarioDetail', () => {
     expect(within(preview).getByRole('alert')).toHaveTextContent('prevent older runs from resuming')
     await user.click(within(preview).getByRole('button', { name: 'Cancel' }))
     expect(mockStartRun).not.toHaveBeenCalled()
-    await user.click(screen.getByRole('button', { name: 'Launch scan' }))
+    await user.click(await screen.findByRole('button', { name: 'Launch scan' }))
     const confirmed = await screen.findByRole('dialog', { name: 'Run preview' })
     await user.click(within(confirmed).getByRole('button', { name: 'Launch scan' }))
     expect(mockStartRun).toHaveBeenCalledTimes(1)

pyrit/converter/insert_punctuation_converter.py

@@ -187,11 +187,10 @@ def _insert_within_words(self, prompt: str, num_insertions: int, punctuation_lis
         # List of chars in the prompt string
         prompt_list = list(prompt)
         rng = self._get_random_generator(stream="punctuation")
-        # Store random indices of prompt_list into insert_indices
-        # If the prompt has only 0 or 1 chars, insert at the end of the prompt
-        insert_indices = (
-            [1] if len(prompt_list) <= num_insertions else rng.sample(range(len(prompt_list) - 1), num_insertions)
-        )
+        # Store random indices of prompt_list into insert_indices. A list of n
+        # characters has n + 1 insertion points (before the first character
+        # through after the last), so every position stays reachable.
+        insert_indices = rng.sample(range(len(prompt_list) + 1), num_insertions)
 
         for index in insert_indices:
             # Insert into prompt_list at the insert_indices with random punctuation from the punctuation_list

tests/unit/converter/test_insert_punctuation_converter.py

@@ -2,6 +2,7 @@
 # Licensed under the MIT license.
 
 import re
+from unittest.mock import patch
 
 import pytest
 
@@ -78,3 +79,31 @@ async def test_punctuation_list_rejects_substrings_of_string_punctuation(punctua
     converter = InsertPunctuationConverter(seed=1)
     with pytest.raises(ValueError, match="Invalid punctuations"):
         await converter.convert_async(prompt="hello big world", punctuation_list=punctuation_list)
+
+
+@pytest.mark.parametrize(
+    ("prompt", "offset"),
+    [("a", 0), ("a", 1), ("abc", 0), ("abc", 1), ("abc", 2), ("abc", 3)],
+)
+async def test_insert_within_words_reaches_every_insertion_position_async(*, prompt: str, offset: int) -> None:
+    converter = InsertPunctuationConverter(between_words=False, word_swap_ratio=1.0, seed=1)
+    with patch.object(converter, "_get_random_generator") as get_rng:
+        rng = get_rng.return_value
+        rng.sample.return_value = [offset]
+        rng.choice.return_value = "!"
+        result = await converter.convert_async(prompt=prompt, punctuation_list=["!"])
+
+    population, count = rng.sample.call_args.args
+    assert list(population) == list(range(len(prompt) + 1))
+    assert count == 1
+    assert result.output_text == f"{prompt[:offset]}!{prompt[offset:]}"
+    assert result.output_type == "text"
+
+
+@pytest.mark.parametrize("prompt", ["", "...", " \t\n"])
+async def test_insert_within_words_without_words_async(prompt: str) -> None:
+    converter = InsertPunctuationConverter(between_words=False, word_swap_ratio=1.0, seed=1)
+    result = await converter.convert_async(prompt=prompt, punctuation_list=["!"])
+
+    assert result.output_text == f"!{prompt}"
+    assert result.output_type == "text"