Merge master into the Gemini and Antigravity branch

Both sides rewrote the doctor: master rebuilt it around ready flags so a
fully local setup stops reading as broken, while this branch taught it
that Google has a key of its own and that the local model has neither a
key nor a program. Kept master's structure and folded the branch in: the
two hosted providers answer for their keys, a CLI for its program, and
the JSON keeps both the provider-aware "key" and master's "ready".

SECRET_KEYS gained groq on master and gemini here; the resolution keeps
all four. _conclude was also changed by both: master stopped matching
stderr wording and asks _API_TROUBLE instead, this branch renamed its
last parameter to the provider's short name so a session is stored under
the name it is read back by; the rename now rides on master's body.
codex_models() and the settings loaders landed beside the new Gemini
ones, so both stay. The new cleanup tests called the CLI fake by its old
name, fake_run, which master had renamed to fake_cli.

tests/__init__.py

@@ -17,6 +17,7 @@ os.environ.setdefault("QT_QPA_PLATFORM", "offscreen")
 _SANDBOX = tempfile.mkdtemp(prefix="dikte-tests-")
 os.environ["XDG_CONFIG_HOME"] = os.path.join(_SANDBOX, "config")
 os.environ["XDG_DATA_HOME"] = os.path.join(_SANDBOX, "data")
+os.environ["XDG_CACHE_HOME"] = os.path.join(_SANDBOX, "cache")
 # Home goes with them: the shortcut file, the applications directory and every
 # macOS path start from it rather than from an XDG variable, and a test run is
 # not allowed to touch the real one.

tests/test_assistant.py

@@ -10,21 +10,28 @@ import io
 import json
 import os
 import subprocess
+import sys
+import threading
 import time
 import unittest
 from unittest import mock
 
 from dikte import assistant
-from tests.support import DikteTest, fake_urlopen, only_these_tools
+from tests.support import (DikteTest, FakeCompleted, fake_urlopen,
+                           only_these_tools)
 
 
 class FakeCli:
-    """A CLI that prints the given events and exits."""
+    """A CLI that prints the given events and exits.
 
-    def __init__(self, events=(), code=0, stderr="", noise=()):
+    Its stderr is not modelled: _stream hands the process a temporary file for
+    that, and a mocked Popen leaves the file empty, which is what a quiet CLI
+    writes anyway.
+    """
+
+    def __init__(self, events=(), code=0, noise=()):
         lines = list(noise) + [json.dumps(event) for event in events]
         self.stdout = io.StringIO("\n".join(lines) + "\n")
-        self.stderr = io.StringIO(stderr)
         self.returncode = code
         self.killed = False
 
@@ -41,6 +48,38 @@ class FakeCli:
         self.killed = True
 
 
+class WedgedCli:
+    """A CLI whose stdout produces nothing, the way a hung process's does.
+
+    Iterating its stdout blocks until the kill arrives, because that is what
+    reading a silent pipe does: only the process ending closes the stream.
+    """
+
+    def __init__(self):
+        self.pid = 4242
+        self.returncode = None
+        self.released = threading.Event()
+        self.stdout = self
+
+    def __iter__(self):
+        return self
+
+    def __next__(self):
+        # The 5 second cap is a safety net for the test itself; the kill is
+        # what is supposed to end the wait.
+        self.released.wait(timeout=5)
+        raise StopIteration
+
+    def poll(self):
+        return self.returncode
+
+    def wait(self, timeout=None):
+        return self.returncode
+
+    def close(self):
+        pass
+
+
 class Provider(DikteTest):
     def test_the_default(self):
         self.assertEqual(assistant.provider(self.config()), "claude")
@@ -247,19 +286,7 @@ class Denials(DikteTest):
         self.assertIn("Write", warning)
 
 
-class SessionMissing(unittest.TestCase):
-    def test_a_session_that_is_gone(self):
-        for text in ("Error: session abc not found",
-                     "No conversation with that id",
-                     "unknown thread: abc"):
-            with self.subTest(text=text):
-                self.assertTrue(assistant._session_missing(text))
-
-    def test_an_unrelated_failure(self):
-        for text in ("", "network unreachable", "session limit exceeded"):
-            with self.subTest(text=text):
-                self.assertFalse(assistant._session_missing(text))
-
+class LastLine(unittest.TestCase):
     def test_the_last_line_is_the_one_worth_showing(self):
         self.assertEqual(assistant.last_line("warning\n\nreal error\n"),
                          "real error")
@@ -302,6 +329,28 @@ class Conclude(DikteTest):
             assistant._conclude(self.found(), 1, "session abc not found",
                                 "abc", "claude")
 
+    def test_the_recovery_no_longer_hangs_on_the_words_the_cli_chose(self):
+        # The complaint used to be matched by substring, which a CLI update or
+        # another language broke. A resumed run that died with nothing to show
+        # is now enough on its own.
+        for stderr in ("Oturum bulunamadı", "something else entirely", ""):
+            with self.subTest(stderr=stderr):
+                with self.assertRaises(assistant._SessionGone):
+                    assistant._conclude(self.found(), 1, stderr, "abc", "Claude")
+
+    def test_api_trouble_on_a_resumed_run_is_not_blamed_on_the_session(self):
+        # A fresh session cannot cure a spent quota, a signed-out CLI or a dead
+        # network: the retry would fail the same way after a second wait, and
+        # the user would lose the conversation thread on top.
+        for stderr in ("Rate limit exceeded",
+                       "You are not logged in. Please run /login.",
+                       "API Error: 401 Unauthorized",
+                       "fetch failed: ECONNREFUSED 127.0.0.1"):
+            with self.subTest(stderr=stderr):
+                with self.assertRaises(assistant.AssistantError) as caught:
+                    assistant._conclude(self.found(), 1, stderr, "abc", "Claude")
+                self.assertIn(stderr, str(caught.exception))
+
     def test_a_session_that_is_gone_only_matters_when_one_was_resumed(self):
         with self.assertRaises(assistant.AssistantError):
             assistant._conclude(self.found(), 1, "session abc not found",
@@ -312,6 +361,11 @@ class Conclude(DikteTest):
                                         "", "claude")
         self.assertEqual(answer, "done")
 
+    def test_an_answer_on_a_resumed_session_is_kept_rather_than_retried(self):
+        answer, _ = assistant._conclude(self.found(answer="done"), 1, "noise",
+                                        "abc", "Claude")
+        self.assertEqual(answer, "done")
+
     def test_a_reported_failure_with_no_answer(self):
         with self.assertRaises(assistant.AssistantError) as caught:
             assistant._conclude(self.found(failure="the model refused"), 0, "",
@@ -324,14 +378,53 @@ class Conclude(DikteTest):
         self.assertIn("Codex", str(caught.exception))
 
 
+class Stream(DikteTest):
+    def test_a_cli_that_floods_stderr_still_finishes(self):
+        # A real subprocess, because the wedge being tested is real plumbing:
+        # with stderr on a pipe nobody drains, 200 KB fills the pipe's buffer,
+        # the child blocks writing it, and the run hangs until the watchdog
+        # timeout. With stderr on a file the run completes at once.
+        script = (
+            "import sys\n"
+            "sys.stderr.write('x' * 200000)\n"
+            "sys.stderr.flush()\n"
+            "print('{\"type\": \"result\", \"result\": \"done\"}')\n"
+        )
+        conf = self.config(assistant_timeout=15)
+        events = []
+        code, stderr = assistant._stream(
+            [sys.executable, "-c", script], conf, events.append, None)
+        self.assertEqual(code, 0)
+        self.assertEqual(len(stderr), 200000)
+        self.assertEqual(events[-1]["result"], "done")
+
+    def test_the_watchdog_takes_the_whole_tree_down_on_timeout(self):
+        proc = WedgedCli()
+
+        def killed(target):
+            # What the real kill does, as far as _stream can see: the process
+            # ends, and its closing stream releases the blocked read.
+            target.returncode = 1
+            target.released.set()
+
+        conf = self.config(assistant_timeout=0)
+        with mock.patch.object(subprocess, "Popen", return_value=proc), \
+                mock.patch.object(assistant, "kill_tree",
+                                  side_effect=killed) as kill:
+            with self.assertRaises(assistant.AssistantError) as caught:
+                assistant._stream(["claude"], conf, lambda event: None, None)
+        kill.assert_called_once_with(proc)
+        self.assertIn("did not finish", str(caught.exception))
+
+
 class AskClaude(DikteTest):
-    def run_ask(self, conf=None, events=None, code=0, stderr="", noise=(),
+    def run_ask(self, conf=None, events=None, code=0, noise=(),
                 session=""):
         conf = conf or self.config()
         proc = FakeCli(events or [
             {"type": "system", "subtype": "init", "session_id": "abc"},
             {"type": "result", "session_id": "abc", "result": "  done  "},
-        ], code=code, stderr=stderr, noise=noise)
+        ], code=code, noise=noise)
         stages = []
         with only_these_tools("claude", "codex"), \
                 mock.patch.object(subprocess, "Popen", return_value=proc) as popen:
@@ -669,6 +762,89 @@ class Ask(DikteTest):
         self.assertEqual(attempts, ["stale-id", ""])
         self.assertEqual(assistant.stored_provider(), "")
 
+    def test_a_resumed_run_that_dies_is_retried_without_the_session_flag(self):
+        # All the way through the stream this time: the first run exits 1 with
+        # no answer and whatever stderr it liked, and the recovery must not
+        # depend on those words.
+        conf = self.config()
+        assistant.write_session("claude", "stale-id")
+        procs = iter([
+            FakeCli(code=1),
+            FakeCli(events=[{"type": "result", "result": "done"}]),
+        ])
+        cmds = []
+
+        def popen(cmd, **kwargs):
+            cmds.append(cmd)
+            return next(procs)
+
+        with only_these_tools("claude"), \
+                mock.patch.object(subprocess, "Popen", side_effect=popen):
+            answer, _ = assistant.ask("hi", conf)
+        self.assertEqual(answer, "done")
+        self.assertEqual(cmds[0][cmds[0].index("--resume") + 1], "stale-id")
+        self.assertNotIn("--resume", cmds[1])
+
+    def test_a_run_that_dies_with_an_answer_in_hand_is_not_retried(self):
+        conf = self.config()
+        assistant.write_session("claude", "stale-id")
+        calls = []
+
+        def popen(cmd, **kwargs):
+            calls.append(cmd)
+            return FakeCli(events=[{"type": "result", "result": "done"}], code=1)
+
+        with only_these_tools("claude"), \
+                mock.patch.object(subprocess, "Popen", side_effect=popen):
+            answer, _ = assistant.ask("hi", conf)
+        self.assertEqual(answer, "done")
+        self.assertEqual(len(calls), 1)
+
+
+class CodexModels(DikteTest):
+    """The model list read off `codex debug models`."""
+
+    CATALOG = {"models": [
+        {"slug": "gpt-6-mini", "visibility": "list", "priority": 9},
+        {"slug": "gpt-6", "visibility": "list", "priority": 1},
+        {"slug": "codex-auto-review", "visibility": "hide", "priority": 3},
+    ]}
+
+    def models(self, reply, code=0):
+        with only_these_tools("codex"), \
+                mock.patch.object(subprocess, "run",
+                                  return_value=FakeCompleted(
+                                      returncode=code, stdout=reply)) as run:
+            found = assistant.codex_models()
+        self.run_call = run
+        return found
+
+    def test_the_catalog_arrives_best_first_without_the_hidden_ones(self):
+        found = self.models(json.dumps(self.CATALOG))
+        self.assertEqual(found, ["gpt-6", "gpt-6-mini"])
+        self.assertEqual(self.run_call.call_args.args[0],
+                         ["codex", "debug", "models"])
+
+    def test_a_codex_that_is_not_installed_is_not_run(self):
+        with only_these_tools(), \
+                mock.patch.object(subprocess, "run") as run:
+            self.assertEqual(assistant.codex_models(), [])
+        run.assert_not_called()
+
+    def test_a_codex_too_old_to_have_the_command(self):
+        self.assertEqual(self.models("error: unknown subcommand", code=2), [])
+
+    def test_a_catalog_that_is_not_what_was_expected(self):
+        self.assertEqual(self.models(json.dumps(["gpt-6"])), [])
+        self.assertEqual(self.models(""), [])
+
+    def test_a_codex_that_hangs_is_given_up_on(self):
+        with only_these_tools("codex"), \
+                mock.patch.object(subprocess, "run",
+                                  side_effect=subprocess.TimeoutExpired(
+                                      ["codex"], 30)):
+            self.assertEqual(assistant.codex_models(), [])
+
 
 if __name__ == "__main__":
     unittest.main()

tests/test_audio.py

@@ -81,6 +81,14 @@ class ChunkLevels(unittest.TestCase):
         self.assertEqual(peak, 1.0)
         self.assertEqual(rms, 1.0)
 
+    def test_the_fast_and_plain_rms_paths_agree(self):
+        """sumprod is a speedup, not a different sum: on a 3.11 machine the
+        loop must land on the same integers."""
+        chunk = tone(0.1)
+        with mock.patch.object(audio, "sumprod", None):
+            plain = audio.chunk_levels(chunk)
+        self.assertEqual(audio.chunk_levels(chunk), plain)
+
 
 class StereoLevels(unittest.TestCase):
     def test_the_channels_are_read_apart(self):
@@ -280,6 +288,48 @@ class _StalledStream:
         self._released.set()
 
 
+class _DribblingStream:
+    """A pipe that never fills a whole chunk in one read, the way an unbuffered
+    pipe hands data over under load."""
+
+    def __init__(self, data, piece):
+        self._data = io.BytesIO(data)
+        self._piece = piece
+
+    def read(self, size):
+        return self._data.read(min(size, self._piece))
+
+
+class _RunSwappingStream:
+    """A pipe whose recorder moves on mid-read, the way a new recording starts
+    while a stale pump is still draining the old one."""
+
+    def __init__(self, data, recorder, swap_at, new_proc):
+        self._data = io.BytesIO(data)
+        self._recorder = recorder
+        self._swap_at = swap_at
+        self._new_proc = new_proc
+        self.reads = 0
+
+    def read(self, size):
+        if self.reads == self._swap_at:
+            self._recorder._run = object()
+            self._recorder._proc = self._new_proc
+        self.reads += 1
+        return self._data.read(size)
+
+
+class _SignalCrashingProcess(FakeProcess):
+    """An ffmpeg that calls being interrupted a failure, the way ffmpeg does."""
+
+    def __init__(self, data, code=255):
+        super().__init__(data)
+        self._code = code
+
+    def poll(self):
+        return None if self._alive else self._code
+
+
 class _HeldStream:
     """A capture that is paused and taken up again partway through, the way a
     key press lands in the middle of a recording rather than between two."""
@@ -307,7 +357,10 @@ class RecordingCommand(OnLinux, DikteTest):
         super().setUp()
         # Whether pw-record takes --raw is read off the installed binary, and
         # what is being tested here is the command rather than the machine the
-        # test is running on. PwRecordRawOption covers the reading itself.
+        # test is running on. PwRecordRawOption covers the reading itself. The
+        # answer is remembered between calls, so it cannot be remembered
+        # between tests.
+        self.enterContext(mock.patch.object(audio, "_PW_RAW", None))
         self.enterContext(mock.patch.object(
             audio, "_pw_record_raw_option", return_value=["--raw"]))
 
@@ -392,11 +445,30 @@ class PwRecordRawOption(DikteTest):
         self.assertEqual(["--raw"], self.option(return_value=FakeCompleted()))
 
 
+class PwRawMemo(OnLinux, DikteTest):
+    """The --raw probe runs once per process, not once per key press."""
+
+    def setUp(self):
+        super().setUp()
+        self.enterContext(mock.patch.object(audio, "_PW_RAW", None))
+
+    def test_the_probe_is_asked_once_and_remembered(self):
+        with only_these_tools("pw-record"), \
+                mock.patch.object(audio, "_pw_record_raw_option",
+                                  return_value=["--raw"]) as probe:
+            first = audio.recording_command()
+            second = audio.recording_command()
+        probe.assert_called_once_with()
+        self.assertIn("--raw", first)
+        self.assertEqual(first, second)
+
+
 class RecorderChain(OnLinux, DikteTest):
     """Start to WAV, with pw-record faked out."""
 
     def setUp(self):
         super().setUp()
+        self.enterContext(mock.patch.object(audio, "_PW_RAW", None))
         self.enterContext(mock.patch.object(
             audio, "_pw_record_raw_option", return_value=["--raw"]))
 
@@ -476,42 +548,121 @@ class RecorderChain(OnLinux, DikteTest):
         self.assertEqual(len(failures), 1)
         self.assertIn("pulseaudio-utils", failures[0])
 
-    def pump(self, data=b"", stderr=b"", stopping=False, cancelled=False):
+    def pump(self, data=b"", stderr=b"", stopping=False, cancelled=False,
+             alive=False):
         """Run the pump in this thread, where a queued signal would need an
         event loop nobody is running here."""
         recorder = audio.Recorder()
         failures = []
+        deaths = []
         recorder.failed.connect(failures.append)
+        recorder.died.connect(lambda: deaths.append(True))
         proc = FakeProcess(data)
-        proc.stderr = io.BytesIO(stderr)
-        proc._alive = False
+        proc._alive = alive
         recorder._proc = proc
-        recorder._max_bytes = 10 ** 9
+        recorder._log = io.BytesIO(stderr)
         recorder._stopping = stopping
         recorder._cancelled = cancelled
-        recorder._pump()
-        return failures
+        recorder._run = run = object()
+        recorder._pump(run, proc, proc.stdout, recorder._buffer,
+                       recorder._rms, 10 ** 9)
+        return failures, deaths
 
     def test_a_recorder_that_died_on_its_own_says_so(self):
         """parec refused the device, or the sound server went away."""
-        failures = self.pump(stderr=b"connection refused\n")
+        failures, _ = self.pump(stderr=b"connection refused\n")
         self.assertEqual(len(failures), 1)
         self.assertIn("connection refused", failures[0])
 
     def test_a_death_with_nothing_on_stderr_still_names_the_exit_code(self):
-        failures = self.pump()
+        failures, _ = self.pump()
         self.assertIn("exit code", failures[0])
 
+    def test_a_death_that_left_no_exit_code_is_not_named_none(self):
+        """A process nobody managed to reap has no code to show, and "exit
+        code None" would only puzzle the person reading it."""
+        failures, _ = self.pump(alive=True)
+        self.assertEqual(len(failures), 1)
+        self.assertNotIn("None", failures[0])
+
     def test_a_recording_we_ended_ourselves_is_not_a_death(self):
         """Otherwise a stray keypress produces two errors, and the first one
         sends the user looking for a broken sound server."""
-        self.assertEqual(self.pump(stopping=True), [])
+        self.assertEqual(self.pump(stopping=True), ([], []))
 
     def test_a_cancelled_recording_is_not_a_death(self):
-        self.assertEqual(self.pump(cancelled=True), [])
+        self.assertEqual(self.pump(cancelled=True), ([], []))
 
-    def test_a_recorder_that_captured_something_first_is_not_a_death(self):
-        self.assertEqual(self.pump(data=silence(0.5)), [])
+    def test_a_capture_that_ends_mid_recording_dies_rather_than_fails(self):
+        """Sound had already arrived, so this is not a broken installation:
+        the app is told the recording died and can rescue what there is."""
+        failures, deaths = self.pump(data=silence(0.5))
+        self.assertEqual(failures, [])
+        self.assertEqual(deaths, [True])
+
+    def test_short_pipe_reads_are_gathered_into_whole_chunks(self):
+        """Every RMS entry must stand for one full chunk, or the silence check
+        weighs a half-filled read as its own stretch of room tone."""
+        half = audio.CHUNK_BYTES // 2
+        data = pcm([1000] * (3 * half // 2))   # three half-chunk reads
+        recorder = audio.Recorder()
+        proc = FakeProcess(b"")
+        proc.stdout = _DribblingStream(data, half)
+        proc._alive = False
+        recorder._proc = proc
+        recorder._log = io.BytesIO(b"")
+        recorder._run = run = object()
+        buffer, rms = bytearray(), []
+        recorder._pump(run, proc, proc.stdout, buffer, rms, 10 ** 9)
+        self.assertEqual(len(buffer), len(data))
+        self.assertEqual(len(rms), 2)   # one whole chunk, then the tail
+
+    def test_a_stale_pump_cannot_touch_the_recording_that_replaced_it(self):
+        """A pump that outlives its join must not meter the next run, stop its
+        process, push audio into its buffer, or speak on its behalf."""
+        recorder = audio.Recorder()
+        levels, failures, deaths = [], [], []
+        recorder.level.connect(levels.append)
+        recorder.failed.connect(failures.append)
+        recorder.died.connect(lambda: deaths.append(True))
+        new_proc = FakeProcess(b"")
+        old_proc = FakeProcess(b"")
+        old_proc.stdout = _RunSwappingStream(tone(0.192), recorder,
+                                             swap_at=1, new_proc=new_proc)
+        recorder._proc = old_proc
+        recorder._log = io.BytesIO(b"")
+        old_run = object()
+        recorder._run = old_run
+        recorder._buffer = bytearray()   # the next recording's buffer
+        old_buffer, old_rms = bytearray(), []
+        recorder._pump(old_run, old_proc, old_proc.stdout, old_buffer, old_rms,
+                       2 * audio.CHUNK_BYTES)
+        # Metered once, then the new run took over: the over-length cutoff hit
+        # on the next chunk and had to stand down instead of stopping a
+        # process that was never its own.
+        self.assertEqual(len(levels), 1)
+        self.assertEqual(len(old_buffer), 2 * audio.CHUNK_BYTES)
+        self.assertEqual(new_proc.signals, [])
+        self.assertEqual(old_proc.signals, [])
+        self.assertEqual(recorder._buffer, bytearray())
+        self.assertEqual((failures, deaths), ([], []))
+
+    def test_a_wav_that_cannot_be_written_is_reported_not_raised(self):
+        recorder = audio.Recorder()
+        results, failures = [], []
+        recorder.stopped.connect(lambda *args: results.append(args))
+        recorder.failed.connect(failures.append)
+        proc = FakeProcess(tone(1.0))
+        with only_these_tools("pw-record"), \
+                mock.patch.object(subprocess, "Popen", return_value=proc), \
+                mock.patch.object(audio, "write_wav",
+                                  side_effect=OSError("disk full")):
+            recorder.start()
+            recorder._thread.join(timeout=5)
+            recorder.stop()
+        self.assertEqual(results, [])
+        self.assertEqual(len(failures), 1)
+        self.assertIn("disk full", failures[0])
 
     def test_a_short_recording_reports_only_that(self):
         _, results, failures, _ = self.record(silence(0.1))
@@ -722,6 +873,42 @@ class MacMeetingRecorder(OnMacOS, DikteTest):
         _, _, _, _, processes, _ = self.record(tone(0.5), tone(0.5))
         self.assertTrue(all(process.signals for process in processes))
 
+    def test_a_stop_we_asked_for_is_not_reported_as_an_ffmpeg_failure(self):
+        """ffmpeg exits 255 when interrupted, and the interruption was our own
+        stop: a meeting ended at once must say "too short", not "ffmpeg → 255"."""
+        path = str(self.path("meeting.wav"))
+        recorder = audio.MeetingRecorder()
+        failed = []
+        recorder.failed.connect(failed.append)
+        processes = [_SignalCrashingProcess(tone(0.1)),
+                     _SignalCrashingProcess(tone(0.1))]
+        with only_these_tools("ffmpeg"), self.devices(), \
+                mock.patch.object(subprocess, "Popen", side_effect=processes):
+            recorder.start(path, "MacBook Pro Microphone", "BlackHole 2ch")
+            recorder._thread.join(timeout=5)
+            recorder.stop()
+        self.assertEqual(len(failed), 1)
+        self.assertIn("0.3", failed[0])
+        self.assertNotIn("255", failed[0])
+
+    def test_an_ffmpeg_that_died_on_its_own_keeps_its_exit_code(self):
+        """A process nobody interrupted has a story to tell, and its code is
+        the only lead the user gets."""
+        path = str(self.path("meeting.wav"))
+        recorder = audio.MeetingRecorder()
+        failed = []
+        recorder.failed.connect(failed.append)
+        dead = _SignalCrashingProcess(tone(0.1))
+        dead._alive = False   # it fell over before stop() reached it
+        processes = [dead, FakeProcess(tone(0.1))]
+        with only_these_tools("ffmpeg"), self.devices(), \
+                mock.patch.object(subprocess, "Popen", side_effect=processes):
+            recorder.start(path, "MacBook Pro Microphone", "BlackHole 2ch")
+            recorder._thread.join(timeout=5)
+            recorder.stop()
+        self.assertEqual(len(failed), 1)
+        self.assertIn("255", failed[0])
+
     def test_a_legacy_numeric_target_fails_before_recording(self):
         recorder = audio.MeetingRecorder()
         failed = []
@@ -953,9 +1140,11 @@ class WindowsDevices(OnWindows, DikteTest):
     def setUp(self):
         super().setUp()
         # The listing is remembered between calls, so that a dictation does not
-        # run ffmpeg of its own. It cannot be remembered between tests.
+        # run ffmpeg of its own. It cannot be remembered between tests, and
+        # neither can the pw-record probe's answer.
         audio._DSHOW_SEEN.clear()
         self.addCleanup(audio._DSHOW_SEEN.clear)
+        self.enterContext(mock.patch.object(audio, "_PW_RAW", None))
 
     @contextlib.contextmanager
     def listing(self, stderr=None, tools=("ffmpeg",)):

tests/test_cleanup.py

@@ -1,9 +1,9 @@
-"""Who cleans the transcript up, and what they are asked.
+"""Who cleans the transcript up, and what they are asked.
 
-The CLIs are faked at subprocess.run: what the tests read is the argument list
-each one is given, where the answer is picked up from, and what happens to the
-chain when the program is missing, slow or unhappy. The OpenRouter path is the
-one that was always there and is checked here only for still being taken.
+The CLIs are faked at subprocess.Popen: what the tests read is the argument
+list each one is given, where the answer is picked up from, and what happens to
+the chain when the program is missing, slow or unhappy. The OpenRouter path is
+the one that was always there and is checked here only for still being taken.
 """
 
 import os
@@ -18,18 +18,26 @@ from tests.support import DikteTest, fake_urlopen, sent_json, url_error
 from tests.test_api import FakeServer, chat_reply
 
 
-def fake_run(stdout="", code=0, stderr="", last_message=""):
-    """Stand in for subprocess.run, writing the file Codex would have written."""
+def fake_cli(stdout="", code=0, stderr="", last_message=""):
+    """Stand in for subprocess.Popen.
+
+    _output hands the process a temporary file for each stream, so the fake
+    writes into those, plus the file Codex would have written on its way out.
+    """
     calls = []
 
-    def run(cmd, **kwargs):
+    def popen(cmd, **kwargs):
         calls.append(cmd)
+        kwargs["stdout"].write(stdout.encode("utf-8"))
+        kwargs["stderr"].write(stderr.encode("utf-8"))
         if last_message and "-o" in cmd:
             with open(cmd[cmd.index("-o") + 1], "w", encoding="utf-8") as fh:
                 fh.write(last_message)
-        return subprocess.CompletedProcess(cmd, code, stdout, stderr)
+        proc = mock.Mock()
+        proc.returncode = code
+        return proc
 
-    return mock.patch.object(subprocess, "run", side_effect=run), calls
+    return mock.patch.object(subprocess, "Popen", side_effect=popen), calls
 
 
 class Provider(DikteTest):
@@ -92,7 +100,7 @@ class OpenRouter(DikteTest):
 
     def test_no_cli_is_started_for_it(self):
         conf = self.config(openrouter_api_key="sk-or-test")
-        patcher, calls = fake_run(stdout="never")
+        patcher, calls = fake_cli(stdout="never")
         with patcher, mock.patch.object(api, "cleanup", return_value="Done."):
             cleanup.run("uh, done", conf, "the rules")
         self.assertEqual(calls, [])
@@ -132,7 +140,7 @@ class GoogleAiStudio(DikteTest):
         self.assertIn("Google AI Studio", str(caught.exception))
 
     def test_no_cli_is_started_for_it(self):
-        patcher, calls = fake_run(stdout="never")
+        patcher, calls = fake_cli(stdout="never")
         with patcher, fake_urlopen(chat_reply("Done.")):
             cleanup.run("uh, done", self.conf, "the rules")
         self.assertEqual(calls, [])
@@ -145,7 +153,7 @@ class ClaudeCode(DikteTest):
         self.patch_attr(cleanup.shutil, "which", lambda name: f"/usr/bin/{name}")
 
     def run_cleanup(self, text="uh, book it", **kwargs):
-        patcher, calls = fake_run(**kwargs)
+        patcher, calls = fake_cli(**kwargs)
         with patcher:
             answer = cleanup.run(text, self.conf, "the rules")
         return answer, calls[0]
@@ -198,14 +206,18 @@ class ClaudeCode(DikteTest):
             self.run_cleanup(stdout="Book it.")
         self.assertIn("claude", str(caught.exception))
 
-    def test_a_run_that_never_ends(self):
-        def run(cmd, **kwargs):
-            raise subprocess.TimeoutExpired(cmd, 180)
+    def test_a_run_that_never_ends_is_killed_with_its_whole_tree(self):
+        def popen(cmd, **kwargs):
+            proc = mock.Mock()
+            proc.wait.side_effect = subprocess.TimeoutExpired(cmd, 180)
+            return proc
 
-        with mock.patch.object(subprocess, "run", side_effect=run):
+        with mock.patch.object(subprocess, "Popen", side_effect=popen), \
+                mock.patch.object(cleanup.assistant, "kill_tree") as kill:
             with self.assertRaises(cleanup.CleanupError) as caught:
                 cleanup.run("uh, book it", self.conf, "the rules")
         self.assertIn("180", str(caught.exception))
+        kill.assert_called_once()
 
 
 class Codex(DikteTest):
@@ -215,7 +227,7 @@ class Codex(DikteTest):
         self.patch_attr(cleanup.shutil, "which", lambda name: f"/usr/bin/{name}")
 
     def run_cleanup(self, text="uh, book it", **kwargs):
-        patcher, calls = fake_run(**kwargs)
+        patcher, calls = fake_cli(**kwargs)
         with patcher:
             answer = cleanup.run(text, self.conf, "the rules")
         return answer, calls[0]
@@ -268,7 +280,7 @@ class Antigravity(DikteTest):
         self.patch_attr(cleanup.shutil, "which", lambda name: f"/usr/bin/{name}")
 
     def run_cleanup(self, text="uh, book it", **kwargs):
-        patcher, calls = fake_run(**kwargs)
+        patcher, calls = fake_cli(**kwargs)
         with patcher:
             answer = cleanup.run(text, self.conf, "the rules")
         return answer, calls[0]
@@ -389,7 +401,8 @@ class Here(DikteTest):
         self.assertIn("out of memory", str(caught.exception))
 
     def test_no_cli_is_started_for_it(self):
-        patcher, calls = fake_run(stdout="never")
+        patcher, calls = fake_cli(stdout="never")
         with patcher, fake_urlopen(chat_reply("Done.")):
             cleanup.run("uh, done", self.conf, "the rules")
         self.assertEqual(calls, [])
+

tests/test_config.py

@@ -8,7 +8,10 @@ config and now shadows the default.
 
 import json
 import os
+import pathlib
 import sys
+import threading
+import time
 import unittest
 from unittest import mock
 
@@ -44,6 +47,21 @@ class Loading(DikteTest):
             conf = cfg.Config()
         self.assertEqual(conf["cleanup_model"], cfg.DEFAULTS["cleanup_model"])
 
+    def test_a_config_that_is_not_json_is_set_aside_as_evidence(self):
+        """Left in place it would be overwritten by the very next save."""
+        cfg.CONFIG_DIR.mkdir(parents=True, exist_ok=True)
+        cfg.CONFIG_FILE.write_text("{not json", encoding="utf-8")
+        broken = cfg.CONFIG_FILE.with_suffix(".json.broken")
+        with mock.patch("builtins.print") as told:
+            conf = cfg.Config()
+        self.assertEqual(broken.read_text(encoding="utf-8"), "{not json")
+        self.assertFalse(cfg.CONFIG_FILE.exists())
+        self.assertIn(str(broken), told.call_args[0][0])
+        conf.save()
+        self.assertEqual(broken.read_text(encoding="utf-8"), "{not json")
+        self.assertEqual(self.read_config_file()["cleanup_model"],
+                         cfg.DEFAULTS["cleanup_model"])
+
     def test_a_config_that_is_json_but_not_an_object(self):
         cfg.CONFIG_DIR.mkdir(parents=True, exist_ok=True)
         cfg.CONFIG_FILE.write_text("[1, 2]", encoding="utf-8")
@@ -113,6 +131,41 @@ class Saving(DikteTest):
         conf.save()
         self.assertEqual(i18n.language(), "tr")
 
+    def test_the_settings_hit_the_disk_before_the_swap(self):
+        """Renaming a file still in the page cache into place makes a power
+        cut a settings wipe, which is what the atomic replace exists to stop."""
+        with mock.patch("os.fsync") as fsync:
+            cfg.Config().save()
+        fsync.assert_called_once()
+
+    def test_a_file_held_briefly_by_a_scanner_does_not_fail_the_save(self):
+        """Antivirus and sync tools on Windows hold a fresh file for a moment,
+        and the rename over it fails until they let go."""
+        attempts = []
+        real_replace = pathlib.Path.replace
+
+        def flaky(path, target):
+            attempts.append(str(target))
+            if len(attempts) < 3:
+                raise PermissionError("held by a scanner")
+            return real_replace(path, target)
+
+        with mock.patch.object(pathlib.Path, "replace", flaky), \
+                mock.patch("time.sleep"):
+            cfg.Config().save()
+        self.assertEqual(len(attempts), 3)
+        self.assertEqual(self.read_config_file()["language"],
+                         cfg.DEFAULTS["language"])
+
+    def test_a_file_held_for_good_still_raises(self):
+        def held(path, target):
+            raise PermissionError("never let go")
+
+        with mock.patch.object(pathlib.Path, "replace", held), \
+                mock.patch("time.sleep"):
+            with self.assertRaises(PermissionError):
+                cfg.Config().save()
+
 
 class Keys(DikteTest):
     def test_a_stored_key_is_used(self):
@@ -352,6 +405,23 @@ class History(DikteTest):
         cfg.delete_history([])
         self.assertEqual(len(cfg.read_history()), 1)
 
+    def test_amending_matches_on_content_and_patches_in_place(self):
+        rows = [self.entry("a"), self.entry("b")]
+        for row in rows:
+            cfg.append_history(row)
+        patched = cfg.amend_history(rows[0], cleanup_error="could not paste")
+        self.assertEqual(patched["cleanup_error"], "could not paste")
+        kept = cfg.read_history()
+        self.assertEqual([row["text"] for row in kept], ["a", "b"])
+        self.assertEqual(kept[0]["cleanup_error"], "could not paste")
+
+    def test_amending_a_row_a_trim_took_away_is_a_no_op(self):
+        row = self.entry("gone")
+        cfg.append_history(row)
+        cfg.clear_history()
+        self.assertIsNone(cfg.amend_history(row, cleanup_error="x"))
+        self.assertEqual(cfg.read_history(), [])
+
     def test_clearing(self):
         cfg.append_history(self.entry("a"))
         cfg.clear_history()
@@ -360,6 +430,40 @@ class History(DikteTest):
     def test_clearing_a_history_that_is_not_there(self):
         cfg.clear_history()   # must not raise
 
+    def test_an_append_during_a_trim_is_not_lost(self):
+        """Trim is read, cut, rewrite; a dictation appended between the read
+        and the rewrite must wait rather than be erased by a rewrite that
+        never saw it. The rewrite is slowed down to hold the race open."""
+        for index in range(10):
+            cfg.append_history(self.entry(str(index)))
+        real_write = cfg._write_history
+        rewriting = threading.Event()
+
+        def slow_write(lines):
+            rewriting.set()
+            time.sleep(0.1)
+            real_write(lines)
+
+        with mock.patch.object(cfg, "_write_history", slow_write):
+            trimmer = threading.Thread(target=cfg.trim_history, args=(3,))
+            trimmer.start()
+            # The trim now holds the lock inside its read-cut-rewrite window.
+            self.assertTrue(rewriting.wait(5))
+            appender = threading.Thread(target=cfg.append_history,
+                                        args=(self.entry("late"),))
+            appender.start()
+            trimmer.join()
+            appender.join()
+        self.assertEqual([row["text"] for row in cfg.read_history()],
+                         ["7", "8", "9", "late"])
+
+    def test_the_rewrite_hits_the_disk_before_the_swap(self):
+        for index in range(5):
+            cfg.append_history(self.entry(str(index)))
+        with mock.patch("os.fsync") as fsync:
+            cfg.trim_history(2)
+        fsync.assert_called_once()
+
 
 class Meetings(DikteTest):
     def entry(self, base, **changes):
@@ -432,6 +536,27 @@ class Meetings(DikteTest):
         cfg.delete_meetings([])
         self.assertEqual(len(cfg.read_meetings()), 1)
 
+    def test_the_index_hits_the_disk_before_the_swap(self):
+        with mock.patch("os.fsync") as fsync:
+            cfg.save_meeting(self.entry("a"))
+        fsync.assert_called_once()
+
+    def test_an_index_held_briefly_by_a_scanner_is_still_written(self):
+        real_replace = pathlib.Path.replace
+        attempts = []
+
+        def flaky(path, target):
+            attempts.append(str(target))
+            if len(attempts) < 3:
+                raise PermissionError("held by a scanner")
+            return real_replace(path, target)
+
+        with mock.patch.object(pathlib.Path, "replace", flaky), \
+                mock.patch("time.sleep"):
+            cfg.save_meeting(self.entry("a"))
+        self.assertEqual(len(attempts), 3)
+        self.assertEqual([row["base"] for row in cfg.read_meetings()], ["a"])
+
 
 class Defaults(unittest.TestCase):
     """The table itself, which every command line and settings tab reads."""

tests/test_filetranscribe.py

@@ -235,6 +235,28 @@ class ChunkSeconds(DikteTest):
         self.assertEqual(ft.chunk_seconds(self.file(ft.UPLOAD_LIMIT * 2), 0), 0.0)
 
 
+class Ffmpeg(DikteTest):
+    """How the converter process is started."""
+
+    def test_its_output_is_read_as_utf8_whatever_the_locale_says(self):
+        """ffmpeg writes UTF-8; read as the locale codepage its messages
+        mojibake, and a byte the codepage cannot place raises from inside
+        communicate itself."""
+        out = str(self.path("out.wav"))
+        with open(out, "wb") as fh:
+            fh.write(b"\x00")
+        proc = mock.Mock()
+        proc.communicate.return_value = ("", "")
+        proc.returncode = 0
+        proc.poll.return_value = 0
+        with mock.patch.object(ft.subprocess, "Popen", return_value=proc) as popen:
+            ft._ffmpeg(["-i", "in.mp4", out], out)
+        kwargs = popen.call_args.kwargs
+        self.assertTrue(kwargs["text"])
+        self.assertEqual(kwargs["encoding"], "utf-8")
+        self.assertEqual(kwargs["errors"], "replace")
+
+
 class Chunks(DikteTest):
     """What each provider is handed, and in how many pieces."""
 

tests/test_ggml.py

@@ -9,6 +9,7 @@ import contextlib
 import hashlib
 import io
 import os
+import pathlib
 import signal
 import sys
 import tarfile
@@ -179,6 +180,21 @@ class Download(Local):
                 ggml.download(item("m.bin", b"x"), target)
         self.assertFalse(target.exists())
 
+    def test_a_target_held_open_keeps_the_finished_download(self):
+        # Windows refuses to replace a file a running server holds open. The
+        # bytes are complete and verified by then, so the .part must survive
+        # the failure rather than being deleted with everything else.
+        data = b"a finished, verified download"
+        target = self.path("data", "models", "m.bin")
+        with fake_urlopen(body(data)):
+            with mock.patch.object(pathlib.Path, "replace",
+                                   side_effect=PermissionError(13, "in use")):
+                with self.assertRaises(ggml.LocalError) as caught:
+                    ggml.download(item("m.bin", data), target)
+        self.assertIn("held open", str(caught.exception))
+        self.assertEqual(target.with_name("m.bin.part").read_bytes(), data)
+        self.assertFalse(target.exists())
+
 
 # --- installing a program -------------------------------------------------
 
@@ -275,6 +291,70 @@ class InstallProgram(Local):
             self.install("whisper-bin-ubuntu-x64.tar.gz", archive=empty)
         self.assertIn("whisper-server", str(caught.exception))
 
+    def test_a_failed_update_leaves_the_working_install_alone(self):
+        # The old install used to be deleted before the new bytes had even
+        # arrived, so a bad download left no local server at all.
+        path, _ = self.install("whisper-bin-ubuntu-x64.tar.gz")
+        with self.assertRaises(ggml.LocalError):
+            self.install("whisper-bin-ubuntu-x64.tar.gz",
+                         archive=b"not an archive at all")
+        self.assertEqual(ggml.installed_program(ggml.WHISPER), path)
+        self.assertTrue(os.path.isfile(path))
+        self.assertEqual(ggml.installed_version(ggml.WHISPER), "v1.9.1")
+        # And the half-made sibling did not linger either.
+        left = list(self.path("data", "bin", "whisper").glob("*.new"))
+        self.assertEqual(left, [])
+
+    def test_an_update_that_never_downloaded_leaves_the_install_alone(self):
+        path, _ = self.install("whisper-bin-ubuntu-x64.tar.gz")
+        listing = self.release("whisper-bin-ubuntu-x64.tar.gz")
+        with serving(listing, self.archive):
+            with mock.patch.object(ggml, "download",
+                                   side_effect=ggml.LocalError("no route")):
+                with self.assertRaises(ggml.LocalError):
+                    ggml.install_program(ggml.WHISPER)
+        self.assertEqual(ggml.installed_program(ggml.WHISPER), path)
+        self.assertTrue(os.path.isfile(path))
+
+    class StubServer:
+        """Owns the installed binary, remembers when it was told to stop."""
+
+        def __init__(self):
+            self.program = ggml.WHISPER
+            self.stops = 0
+            self.new_version_was_ready = False
+
+        def settings(self):
+            return {"binary": ""}
+
+        def stop(self):
+            self.stops += 1
+            self.new_version_was_ready = any(
+                (ggml.BIN_DIR / "whisper").glob("*.new"))
+
+    def test_the_running_server_is_stopped_only_for_the_swap(self):
+        # The outage is the swap, not the transfer: the server keeps answering
+        # through a download that can take minutes, and is stopped only once
+        # the replacement is unpacked next door and known to be whole.
+        self.install("whisper-bin-ubuntu-x64.tar.gz")
+        server = self.StubServer()
+        with mock.patch.object(ggml, "SERVERS", (server,)):
+            self.install("whisper-bin-ubuntu-x64.tar.gz")
+        self.assertEqual(server.stops, 1)
+        self.assertTrue(server.new_version_was_ready)
+
+    def test_a_download_that_fails_never_stops_the_server(self):
+        self.install("whisper-bin-ubuntu-x64.tar.gz")
+        server = self.StubServer()
+        listing = self.release("whisper-bin-ubuntu-x64.tar.gz")
+        with mock.patch.object(ggml, "SERVERS", (server,)):
+            with serving(listing, self.archive):
+                with mock.patch.object(ggml, "download",
+                                       side_effect=ggml.LocalError("no route")):
+                    with self.assertRaises(ggml.LocalError):
+                        ggml.install_program(ggml.WHISPER)
+        self.assertEqual(server.stops, 0)
+
 
     def test_a_release_without_a_published_checksum_is_refused(self):
         # GitHub did not always publish one, and whisper.cpp v1.8.0 and older
@@ -456,12 +536,14 @@ STAND_IN = textwrap.dedent("""
     def opt(name, default=""):
         return args[args.index(name) + 1] if name in args else default
 
+    time.sleep(float(opt("--wait", "0")))
+
+    # After the sleep, so that --wait plus --die is a program that runs for a
+    # while and then crashes, the way a bad model dies mid-load.
     if "--die" in args:
         print("could not load model: no such file")
         sys.exit(2)
 
-    time.sleep(float(opt("--wait", "0")))
-
     started = time.monotonic()
     healthy_after = float(opt("--healthy-after", "0"))
 
@@ -539,6 +621,14 @@ class Servers(Local):
         self.assertTrue(server.running)
         self.assertTrue(second)
 
+    def count_ports(self):
+        """Record every port handed to a launch, one per attempt."""
+        ports = []
+        real = ggml._free_port
+        self.patch_attr(ggml, "_free_port",
+                        lambda: ports.append(real()) or ports[-1])
+        return ports
+
     def test_a_program_that_dies_reports_what_it_printed(self):
         server = self.server(extra=["--die"])
         with self.assertRaises(ggml.LocalError) as caught:
@@ -546,6 +636,46 @@ class Servers(Local):
         self.assertIn("no such file", str(caught.exception))
         self.assertFalse(server.running)
 
+    def test_an_early_death_that_never_listened_is_retried(self):
+        # A child that loses the bind race fails and exits at once, and which
+        # port was lost cannot be told from the log: the shape of the failure,
+        # not its wording, is what earns another port.
+        ports = self.count_ports()
+        server = self.server(extra=["--die"])
+        with self.assertRaises(ggml.LocalError):
+            server.serve()
+        self.assertEqual(len(ports), 3)
+
+    def test_a_late_crash_is_not_retried(self):
+        # A program that ran for a while before dying was not a bind race: it
+        # would die the same way on any port.
+        self.patch_attr(ggml, "EARLY_EXIT_WINDOW", 0.2)
+        ports = self.count_ports()
+        server = self.server(extra=["--wait", "0.5", "--die"])
+        with self.assertRaises(ggml.LocalError) as caught:
+            server.serve()
+        self.assertEqual(len(ports), 1)
+        self.assertIn("no such file", str(caught.exception))
+
+    def test_an_open_port_with_a_dead_child_is_not_ready(self):
+        # Another process winning the bind race leaves the port open while our
+        # child exits: the open port alone must not be read as ready.
+        polls = iter([None, 2])
+        proc = mock.Mock()
+        proc.poll = lambda: next(polls)
+        self.patch_attr(ggml, "_listening", lambda port: True)
+        reason, listened = self.server()._wait_ready(proc, 1)
+        self.assertEqual(reason, "exited")
+        self.assertFalse(listened)
+
+    def test_an_open_port_with_a_child_that_outlives_it_a_beat_is_ready(self):
+        proc = mock.Mock()
+        proc.poll = lambda: None
+        self.patch_attr(ggml, "_listening", lambda port: True)
+        reason, listened = self.server()._wait_ready(proc, 1)
+        self.assertEqual(reason, "ready")
+        self.assertTrue(listened)
+
     def test_a_model_that_is_still_loading_is_not_ready_yet(self):
         # llama binds its port first and answers /health with 503 until the
         # model is in memory, so the open port on its own is not the signal.
@@ -614,6 +744,72 @@ class Servers(Local):
     def test_no_pid_file_is_nothing_to_sweep(self):
         self.assertFalse(self.server().sweep())
 
+    def test_an_unverifiable_pid_is_kept_for_a_later_sweep(self):
+        # Could not be checked is not the same as known stale: dropping the
+        # file here would lose track of a server that may still hold a model.
+        server = self.server()
+        server._remember(4242)
+        with mock.patch.object(server, "_is_ours", return_value=None):
+            self.assertFalse(server.sweep())
+        self.assertTrue(server._pid_file().exists())
+
+    def test_a_pid_known_stale_is_forgotten(self):
+        server = self.server()
+        server._remember(4242)
+        with mock.patch.object(server, "_is_ours", return_value=False):
+            self.assertFalse(server.sweep())
+        self.assertFalse(server._pid_file().exists())
+
+    def test_the_pid_file_goes_once_the_kill_was_attempted(self):
+        server = self.server()
+        server._remember(4242)
+        with mock.patch.object(server, "_is_ours", return_value=True):
+            with mock.patch.object(ggml.os, "kill") as kill:
+                self.assertTrue(server.sweep())
+        kill.assert_called_once_with(4242, signal.SIGTERM)
+        self.assertFalse(server._pid_file().exists())
+
+    def test_forget_leaves_a_pid_file_that_is_no_longer_ours(self):
+        server = self.server()
+        server._remember(111)
+        # A Dikte started after us wrote its own server's pid over ours;
+        # removing the file would hide that server from every future sweep.
+        server._pid_file().write_text("222")
+        server._forget()
+        self.assertEqual(server._pid_file().read_text(), "222")
+
+    def test_forget_removes_the_pid_it_remembered(self):
+        server = self.server()
+        server._remember(111)
+        server._forget()
+        self.assertFalse(server._pid_file().exists())
+
+    def test_stop_waits_out_a_start_in_flight_and_kills_it(self):
+        # stop_all on quit must not slide past a launch that is mid-load: the
+        # child would then survive Dikte with nothing left that knows its pid.
+        children = []
+        real = ggml.subprocess.Popen
+
+        def popen(*args, **kwargs):
+            proc = real(*args, **kwargs)
+            children.append(proc)
+            return proc
+
+        self.patch_attr(ggml.subprocess, "Popen", popen)
+        server = self.server(extra=["--wait", "0.5"])
+        thread = threading.Thread(target=server.serve)
+        thread.start()
+        try:
+            deadline = time.monotonic() + 10
+            while not children and time.monotonic() < deadline:
+                time.sleep(0.01)     # until the launch is truly in flight
+            server.stop()
+        finally:
+            thread.join(timeout=10)
+        self.assertEqual(len(children), 1)
+        self.assertIsNotNone(children[0].poll())
+        self.assertFalse(server.running)
+
     def test_a_start_that_goes_wrong_takes_its_process_with_it(self):
         started = []
 
@@ -752,6 +948,10 @@ class InstallOnWindows(Local):
         super().setUp()
         self.patch_attr(sys, "platform", "win32")
         self.patch_attr(ggml, "_arch", lambda: "x64")
+        # shutil.which cannot be allowed through to the real one: standing on
+        # win32 from another system, Python 3.12's Windows branch of which()
+        # reaches for the nt module that is not there.
+        self.enterContext(mock.patch("shutil.which", return_value=None))
         self.archive = zipball({
             "Release/whisper-server.exe": b"MZ not really a program",
             "Release/whisper.dll": b"not really a library",
@@ -785,3 +985,43 @@ class InstallOnWindows(Local):
             with self.assertRaises(ggml.LocalError) as caught:
                 ggml.install_program(ggml.WHISPER)
         self.assertIn("this machine", str(caught.exception))
+
+
+class WindowsOwnership(Local):
+    """Sweeping on Windows goes by the executable's full path, never its name:
+    the base name alone is anyone's whisper-server.exe, and this answer decides
+    what gets killed."""
+
+    def setUp(self):
+        super().setUp()
+        self.patch_attr(sys, "platform", "win32")
+        # See InstallOnWindows: the real which() on win32 wants the nt module.
+        self.enterContext(mock.patch("shutil.which", return_value=None))
+        self.made = ggml.Server(ggml.WHISPER, lambda values: [], {"binary": ""})
+
+    def image(self, path):
+        self.patch_attr(ggml, "_win_image_name", lambda pid: path)
+
+    def test_a_binary_under_our_bin_directory_is_ours(self):
+        self.image(str(ggml.BIN_DIR / "whisper" / "v1.9.1" / "whisper-server.exe"))
+        self.assertIs(self.made._is_ours(1234), True)
+
+    def test_the_configured_binary_is_ours_wherever_it_lives(self):
+        mine = self.path("elsewhere", "whisper-server.exe")
+        mine.parent.mkdir(parents=True)
+        mine.write_bytes(b"MZ")
+        mine.chmod(0o755)
+        self.made._settings["binary"] = str(mine)
+        self.image(str(mine.resolve()))
+        self.assertIs(self.made._is_ours(1234), True)
+
+    def test_the_same_name_somewhere_else_is_not_ours(self):
+        self.image(str(self.path("theirs", "whisper-server.exe")))
+        with mock.patch("shutil.which", return_value=None):
+            self.assertIs(self.made._is_ours(1234), False)
+
+    def test_a_process_that_cannot_be_read_is_no_verdict(self):
+        # OpenProcess answering nothing covers both "gone" and "not readable
+        # from here", and only one of those makes the pid file safe to drop.
+        self.image("")
+        self.assertIsNone(self.made._is_ours(1234))

tests/test_hub.py

@@ -3,6 +3,7 @@
 import json
 
 from dikte import hub
+from dikte import paths
 from tests.support import DikteTest, fake_urlopen, http_error, url_error
 
 RELEASE = {
@@ -165,6 +166,15 @@ def os_utime(path):
     os.utime(path, (old, old))
 
 
+class CacheLocation(DikteTest):
+    """Resolved at import, like every other path constant."""
+
+    def test_the_cache_lives_in_the_system_cache_directory(self):
+        # One answer for both, the same way ggml and config share DATA_DIR:
+        # hub asked paths once, at import, and kept what it was told.
+        self.assertEqual(hub.CACHE_DIR, paths.cache_dir())
+
+
 class CacheOnDisk(DikteTest):
     def setUp(self):
         super().setUp()

tests/test_integrate.py

@@ -475,6 +475,17 @@ class Windows(unittest.TestCase):
         self.installed = pathlib.Path(self.tmp.name).resolve()
         self.app = self.installed / "Dikte.exe"
         self.app.write_text("")
+        # APPDATA pointed into the sandbox, so that the Startup folder these
+        # tests delete from is never the machine's own.
+        appdata = mock.patch.dict(os.environ,
+                                  {"APPDATA": str(self.installed / "Roaming")})
+        appdata.start()
+        self.addCleanup(appdata.stop)
+
+    def startup_shortcut(self):
+        """Where install.ps1 -Autostart puts a checkout's sign-in entry."""
+        return (self.installed / "Roaming" / "Microsoft" / "Windows"
+                / "Start Menu" / "Programs" / "Startup" / "Dikte.lnk")
 
     def _write(self, command):
         self.value = command
@@ -510,6 +521,45 @@ class Windows(unittest.TestCase):
         self.assertEqual(len(self.install()), 1)
         self.assertEqual(self.value, f'"{self.app}"')
 
+    def test_an_entry_for_another_working_install_is_left_alone(self):
+        """The same courtesy the Linux half pays another menu entry: an entry
+        naming an executable that still exists is an installation that still
+        works, and a start of this one has no business redirecting it."""
+        other = self.installed / "Elsewhere" / "Dikte.exe"
+        other.parent.mkdir()
+        other.write_text("")
+        self.value = f'"{other}"'
+        self.assertEqual(self.install(), [])
+        self.assertEqual(self.value, f'"{other}"')
+
+    def test_asking_outright_overrules_a_working_other_install(self):
+        other = self.installed / "Elsewhere" / "Dikte.exe"
+        other.parent.mkdir()
+        other.write_text("")
+        self.value = f'"{other}"'
+        self.assertEqual(self.install(force=True), [integrate._run_entry_name()])
+        self.assertEqual(self.value, f'"{self.app}"')
+
+    def test_typing_it_sweeps_away_a_checkout_startup_shortcut(self):
+        """install.ps1 -Autostart writes it, the Run value replaces it, and
+        both left in place would be two Diktes at every sign-in."""
+        shortcut = self.startup_shortcut()
+        shortcut.parent.mkdir(parents=True)
+        shortcut.write_text("")
+        changed = self.install(force=True)
+        self.assertIn(shortcut, changed)
+        self.assertFalse(shortcut.exists())
+
+    def test_a_start_leaves_a_checkout_startup_shortcut_alone(self):
+        """The silent call on every start has not been asked to move the
+        machine off its checkout."""
+        shortcut = self.startup_shortcut()
+        shortcut.parent.mkdir(parents=True)
+        shortcut.write_text("")
+        self.value = f'"{self.app}"'
+        self.assertEqual(self.install(), [])
+        self.assertTrue(shortcut.exists())
+
     def test_running_it_again_changes_nothing(self):
         self.install(force=True)
         self.assertEqual(self.install(), [])
@@ -521,6 +571,31 @@ class Windows(unittest.TestCase):
         self.assertEqual(self.remove(), [])
 
 
+class WindowedExecutable(unittest.TestCase):
+    """The windowed executable, looked up beside whichever one is running.
+
+    Beside rather than at a known place: the setup lays both executables into
+    one directory wherever that directory was put, so either can find the
+    other without knowing where the install is.
+    """
+
+    def setUp(self):
+        self.tmp = tempfile.TemporaryDirectory()
+        self.addCleanup(self.tmp.cleanup)
+        self.installed = pathlib.Path(self.tmp.name).resolve()
+
+    def test_found_beside_the_named_executable(self):
+        windowed = self.installed / "Dikte.exe"
+        windowed.write_text("")
+        self.assertEqual(
+            integrate.windowed_executable(str(self.installed / "dikte-cli.exe")),
+            windowed)
+
+    def test_none_when_no_setup_installed_one(self):
+        self.assertIsNone(
+            integrate.windowed_executable(str(self.installed / "dikte-cli.exe")))
+
+
 class WindowsExecutableNames(unittest.TestCase):
     """The two Windows executables, read out of the files that name them.
 

tests/test_ipc.py

@@ -14,6 +14,7 @@ import unittest
 from unittest import mock
 
 from dikte import ipc
+from tests.support import DikteTest
 
 
 class FakeSocket:
@@ -185,5 +186,74 @@ class Send(unittest.TestCase):
         self.assertTrue(sock.disconnected)
 
 
+class AlreadyServing(unittest.TestCase):
+    """The single-instance check, which listen() cannot be: a Windows pipe
+    takes a second server on the same name rather than refusing it."""
+
+    def probe(self, socket):
+        with mock.patch.object(ipc, "QLocalSocket", return_value=socket):
+            return ipc.already_serving()
+
+    def test_nothing_running_means_go_ahead(self):
+        self.assertFalse(self.probe(FakeSocket(connected=False)))
+
+    def test_an_answer_means_yield(self):
+        self.assertTrue(self.probe(FakeSocket(reply=b'{"ok": true}\n')))
+
+    def test_the_probe_has_no_side_effect(self):
+        """A probe that opened a window would open it during the relaunch a
+        slow instance provokes, on top of the verb being forwarded."""
+        sock = FakeSocket(reply=b'{"ok": true}\n')
+        self.probe(sock)
+        self.assertEqual(sock.written.decode("utf-8").strip(), "status")
+
+    def test_an_instance_too_old_to_answer_still_counts_as_running(self):
+        self.assertTrue(self.probe(FakeSocket(reply=b"")))
+
+
+class InstanceLock(DikteTest):
+    def setUp(self):
+        super().setUp()
+        # The lock derives its home from paths, which DikteTest's cfg patches
+        # do not cover; without this the test would write into the real one.
+        from dikte import paths
+        self.patch_attr(paths, "DATA_DIR", self.path("data"))
+
+    def test_one_holder_at_a_time(self):
+        first = ipc.instance_lock()
+        self.assertIsNotNone(first)
+        self.assertTrue(first.tryLock(0))
+        second = ipc.instance_lock()
+        self.assertFalse(second.tryLock(0))
+        first.unlock()
+        self.assertTrue(second.tryLock(0))
+        second.unlock()
+
+    def test_the_lock_lives_in_the_data_directory(self):
+        from dikte import paths
+        lock = ipc.instance_lock()
+        self.assertTrue(lock.tryLock(0))
+        self.assertTrue((paths.DATA_DIR / "dikte.lock").exists())
+        lock.unlock()
+
+
+class Respawn(unittest.TestCase):
+    def test_windows_starts_a_detached_process_and_returns(self):
+        with mock.patch.object(sys, "platform", "win32"), \
+                mock.patch.object(ipc, "launcher", return_value=["py", "x"]), \
+                mock.patch.object(ipc.subprocess, "Popen") as popen:
+            ipc.respawn(["--gui"])
+        self.assertEqual(popen.call_args.args[0], ["py", "x", "--gui"])
+        self.assertEqual(popen.call_args.kwargs["creationflags"],
+                         0x00000008 | 0x00000200)
+
+    def test_everywhere_else_the_process_is_replaced(self):
+        with mock.patch.object(sys, "platform", "linux"), \
+                mock.patch.object(ipc, "launcher", return_value=["py", "x"]), \
+                mock.patch.object(ipc.os, "execv") as execv:
+            ipc.respawn(["toggle", "--gui"])
+        execv.assert_called_once_with("py", ["py", "x", "toggle", "--gui"])
+
+
 if __name__ == "__main__":
     unittest.main()

tests/test_paste.py

@@ -444,6 +444,45 @@ class MacOS(ClipboardContract, DikteTest):
         self.assertFalse(paste.paste_ready())
 
 
+class MacPasteGoesWhereTheDictationStarted(MacOS):
+    """The keys land in the frontmost window, so the front is what decides
+    where a transcript ends up."""
+
+    def setUp(self):
+        super().setUp()
+        from dikte import mac_window
+        self.mac_window = mac_window
+        self.activated = []
+        self.patch_attr(mac_window, "activate", self.activated.append)
+
+    def frontmost(self, dikte_is):
+        self.patch_attr(self.mac_window, "is_frontmost", lambda: dikte_is)
+
+    def test_a_dikte_that_took_the_front_hands_it_back_before_pressing(self):
+        self.frontmost(True)
+        paste.press("cmd+v", focus=4242)
+        self.assertEqual(self.activated, [4242])
+        self.assertEqual([event for _, event in self.api.posted], [1001, 1002])
+
+    def test_another_application_in_front_is_where_the_user_went_and_is_left(self):
+        self.frontmost(False)
+        paste.press("cmd+v", focus=4242)
+        self.assertEqual(self.activated, [])
+
+    def test_a_run_that_remembered_nobody_asks_nothing(self):
+        self.frontmost(True)
+        paste.press("cmd+v")
+        self.assertEqual(self.activated, [])
+
+    def test_the_front_is_handed_back_only_once_macos_trusts_dikte(self):
+        """Pulling the user out of their window and then failing to type would
+        be the worst of both."""
+        self.frontmost(True)
+        self.api.trusted = False
+        with self.assertRaises(paste.PasteError):
+            paste.press("cmd+v", focus=4242)
+        self.assertEqual(self.activated, [])
+
 class MacClipboardSnapshot(DikteTest):
     def test_every_native_type_is_restored_and_the_files_are_removed(self):
         directory = tempfile.mkdtemp(prefix="dikte-test-clipboard-")

tests/test_paths.py

@@ -61,6 +61,34 @@ class Directories(unittest.TestCase):
         self.assertTrue(data_dir.as_posix().endswith("/AppData/Local/Dikte"))
 
 
+class CacheDir(unittest.TestCase):
+    """The third place: files whose whole point is that they can be lost."""
+
+    def test_linux_follows_xdg(self):
+        with mock.patch.dict(os.environ, {"XDG_CACHE_HOME": "/k"}):
+            self.assertEqual(paths.cache_dir("linux").as_posix(), "/k/dikte")
+
+    def test_linux_without_the_variable_set(self):
+        with mock.patch.dict(os.environ, {}, clear=True):
+            self.assertTrue(paths.cache_dir("linux").as_posix()
+                            .endswith("/.cache/dikte"))
+
+    def test_a_mac_caches_under_library_caches(self):
+        """Where Time Machine already knows not to look."""
+        self.assertTrue(paths.cache_dir("darwin").as_posix()
+                        .endswith("/Library/Caches/Dikte"))
+
+    def test_windows_caches_outside_the_roaming_profile(self):
+        with mock.patch.dict(os.environ, {"LOCALAPPDATA": "C:/local"}):
+            self.assertEqual(paths.cache_dir("win32").as_posix(),
+                             "C:/local/Dikte/cache")
+
+    def test_windows_without_the_variable_set(self):
+        with mock.patch.dict(os.environ, {}, clear=True):
+            self.assertTrue(paths.cache_dir("win32").as_posix()
+                            .endswith("/AppData/Local/Dikte/cache"))
+
+
 class OnePlace(unittest.TestCase):
     """The programs and the models go where everything else goes.
 

tests/test_ui.py

@@ -137,6 +137,8 @@ class Settings(DikteTest):
                                             "_load_models"))
         self.enterContext(mock.patch.object(settings_ui.SettingsWindow,
                                             "_load_transcribe_models"))
+        self.enterContext(mock.patch.object(settings_ui.SettingsWindow,
+                                            "_load_codex_models"))
         # The local model boxes fetch their own list the moment they are shown,
         # from a thread, which is nobody's test failing but a real request.
         self.enterContext(mock.patch.object(settings_ui.LocalModelBox,
@@ -244,6 +246,21 @@ class Settings(DikteTest):
             with self.subTest(key=key):
                 self.assertEqual(stored[key], value)
 
+    def test_only_a_save_that_changed_the_language_says_so(self):
+        # The owner answers language_changed by replacing the window, so a
+        # save that left the language alone must keep quiet.
+        i18n = settings_ui.i18n
+        self.addCleanup(i18n.set_language, i18n.language())
+        window = self.window(cfg.Config())
+        heard = []
+        window.language_changed.connect(lambda: heard.append(True))
+        window._save()
+        self.assertEqual(heard, [])
+        other = "en" if i18n.language() == "tr" else "tr"
+        window._select_data(window.ui_language, other)
+        window._save()
+        self.assertEqual(heard, [True])
+
     def test_the_model_box_on_screen_belongs_to_whoever_cleans_up(self):
         """An OpenRouter id and a Claude alias are not the same field."""
         window = self.window(cfg.Config())
@@ -258,6 +275,19 @@ class Settings(DikteTest):
                 self.assertEqual(shown, [provider])
                 self.assertFalse(box.isHidden())
 
+    def test_codex_answering_refills_both_of_its_boxes(self):
+        """The list Codex gave replaces the built-in one, in both places, and
+        neither loses what was already picked."""
+        conf = self.config(cleanup_codex_model="my-own-model")
+        window = self.window(conf)
+        window._on_codex_models_loaded(["gpt-6", "gpt-6-mini"])
+        for combo in (window.cleanup_codex_model, window.assistant_codex_model):
+            with self.subTest(combo=combo.objectName() or "combo"):
+                offered = [combo.itemText(i) for i in range(combo.count())]
+                self.assertEqual(offered[1:], ["gpt-6", "gpt-6-mini"])
+        self.assertEqual(window.cleanup_codex_model.currentText(),
+                         "my-own-model")
+
     def test_the_update_line_names_the_version_that_is_running(self):
         window = self.window(cfg.Config())
         self.assertIn(settings_ui.__version__, window.update_status.text())
@@ -522,6 +552,321 @@ class KdeSettings(Settings):
         self.assertFalse(window.evdev_enabled.isHidden())
 
 
+
+
+class FakeAppKit:
+    """The Objective-C runtime, answering rather than being asked.
+
+    Stands in for what mac_window._appkit() loads, so what the indicator's
+    window would have been sent can be read off `told` on any machine. The
+    selectors come back as the names they were registered under, which is what
+    lets the tests below name the message they mean.
+    """
+
+    def __init__(self, window=4242, panel=True, mask=0xe):
+        self.objc = self
+        self.window, self.panel, self.mask = window, panel, mask
+        self.told = {}
+
+    def objc_getClass(self, name):
+        return 1 if name == b"NSPanel" else 0
+
+    def selector(self, name):
+        return name.decode()
+
+    def shared(self, _class_name, _selector):
+        return 1
+
+    def ask(self, _to, selector):
+        return self.window if selector == "window" else 0
+
+    def ask_unsigned(self, _to, _selector):
+        return self.mask
+
+    def ask_of_class(self, _to, _selector, _klass):
+        return self.panel
+
+    def tell_bool(self, _to, selector, value):
+        self.told[selector] = value
+
+    tell_unsigned = tell_bool
+
+
+class MacIndicatorWindow(DikteTest):
+    """Which of the three AppKit settings the indicator's window is sent.
+
+    The nonactivating bit is the one worth a test of its own: it is legal on an
+    NSPanel and nowhere else, and sending it to a plain NSWindow raises an
+    Objective-C exception, which through ctypes is not something Python can
+    catch. It takes the whole process down. `_is_panel` is the only thing
+    standing between the two, so what it answers has to decide what is sent.
+    """
+
+    def setUp(self):
+        super().setUp()
+        from dikte import mac_window
+        self.mac_window = mac_window
+        self.patch_attr(mac_window.QGuiApplication, "platformName",
+                        staticmethod(lambda: "cocoa"))
+
+    def told(self, **kwargs):
+        """What keep_on_screen sends an indicator, against this AppKit."""
+        appkit = FakeAppKit(**kwargs)
+        self.patch_attr(self.mac_window, "_appkit", lambda: appkit)
+        widget = overlay_module.Overlay()
+        self.addCleanup(widget.deleteLater)
+        self.addCleanup(widget.close)
+        self.answered = self.mac_window.keep_on_screen(widget)
+        return appkit.told
+
+    def test_a_window_that_is_not_a_panel_is_never_sent_the_style_mask(self):
+        """The message that would kill the process. The other two still go."""
+        told = self.told(panel=False)
+        self.assertTrue(self.answered)
+        self.assertNotIn("setStyleMask:", told)
+        self.assertIs(told["setHidesOnDeactivate:"], False)
+        self.assertEqual(told["setCollectionBehavior:"],
+                         self.mac_window.BEHAVIOUR)
+
+    def test_a_panel_without_the_bit_is_sent_the_mask_with_it_added(self):
+        told = self.told(panel=True, mask=0xe)
+        self.assertEqual(told["setStyleMask:"],
+                         0xe | self.mac_window.NONACTIVATING_PANEL)
+
+    def test_a_panel_that_already_has_the_bit_is_left_alone(self):
+        """Every dictation runs this again, and a mask Cocoa did not need is a
+        window it rebuilds underneath the indicator."""
+        told = self.told(panel=True,
+                         mask=0xe | self.mac_window.NONACTIVATING_PANEL)
+        self.assertNotIn("setStyleMask:", told)
+
+    def test_a_window_the_view_does_not_have_yet_is_not_messaged(self):
+        self.assertEqual(self.told(window=0), {})
+        self.assertFalse(self.answered)
+
+    def test_nothing_is_sent_anywhere_but_cocoa(self):
+        """Every other platform hands out a winId that means something else
+        entirely, and messaging it crashes the run."""
+        self.patch_attr(self.mac_window.QGuiApplication, "platformName",
+                        staticmethod(lambda: "offscreen"))
+        self.assertEqual(self.told(), {})
+        self.assertFalse(self.answered)
+
+
+class GivingTheFrontBack(DikteTest):
+    """Opening the microphone brings Dikte to the front, and the window the
+    user was dictating into goes inactive with the caret in it. Nothing can be
+    asked of the capture session, so the front is put back afterwards.
+
+    The watch is driven by hand here: what matters is what it decides, not how
+    long Qt takes to tick.
+    """
+
+    def setUp(self):
+        super().setUp()
+        from dikte import app as dikte_module
+        from dikte import mac_window
+        self.dikte = dikte_module
+        self.activated = []
+        self.patch_attr(mac_window, "activate", self.activate)
+        self.ticks = []
+        outer = self
+
+        class FakeTimer:
+            """Records what it was asked to do and hands over the tick."""
+
+            def __init__(self, _parent):
+                self.interval = None
+                self.running = False
+                outer.ticks.append(self)
+
+            def setInterval(self, milliseconds):
+                self.interval = milliseconds
+
+            def start(self):
+                self.running = True
+
+            def stop(self):
+                self.running = False
+
+            @property
+            def timeout(self):
+                return self
+
+            def connect(self, slot):
+                self.tick = slot
+
+        self.patch_attr(dikte_module, "QTimer", FakeTimer)
+
+        class BareDikte:
+            """As much of the application as this one method touches."""
+
+            app = None
+            _front_watch = None
+            _the_front = dikte_module.Dikte._the_front
+            _give_the_front_back = dikte_module.Dikte._give_the_front_back
+            _stop_watching_the_front = dikte_module.Dikte._stop_watching_the_front
+
+        self.bare = BareDikte
+
+    def activate(self, pid):
+        self.activated.append(pid)
+        return True
+
+    def watching(self, was_in_front, dikte_in_front, on=None):
+        from dikte import mac_window
+        self.patch_attr(mac_window, "is_frontmost", lambda: dikte_in_front)
+        dikte = on if on is not None else self.bare()
+        dikte._give_the_front_back(was_in_front)
+        return self.ticks[-1] if self.ticks else None
+
+    def test_the_front_goes_back_to_whoever_had_it(self):
+        watch = self.watching(4242, dikte_in_front=True)
+        watch.tick()
+        self.assertEqual(self.activated, [4242])
+        self.assertTrue(watch.running)    # accepted is not the same as landed
+        self.patch_attr(self.mac_window_module(), "is_frontmost", lambda: False)
+        watch.tick()
+        self.assertFalse(watch.running)
+
+    def test_an_accepted_restore_is_not_sent_again_while_it_is_landing(self):
+        watch = self.watching(4242, dikte_in_front=True)
+        watch.tick()
+        watch.tick()
+        self.assertEqual(self.activated, [4242])
+        self.assertTrue(watch.running)
+
+    def test_an_accepted_restore_that_never_lands_still_times_out(self):
+        watch = self.watching(4242, dikte_in_front=True)
+        watch.tick()
+        with mock.patch.object(self.dikte.time, "monotonic",
+                               return_value=self.dikte.time.monotonic() + 60):
+            watch.tick()
+        self.assertFalse(watch.running)
+        self.assertEqual(self.activated, [4242])
+
+    def test_a_restore_the_system_refused_is_retried(self):
+        from dikte import mac_window
+        self.patch_attr(mac_window, "activate",
+                        lambda pid: self.activated.append(pid) or False)
+        watch = self.watching(4242, dikte_in_front=True)
+        watch.tick()
+        watch.tick()
+        self.assertEqual(self.activated, [4242, 4242])
+        self.assertTrue(watch.running)
+
+    def test_a_front_that_was_never_taken_is_left_where_it_is(self):
+        """The microphone does not always take it, and pulling an application
+        forward that is already there is one flicker for nothing."""
+        watch = self.watching(4242, dikte_in_front=False)
+        watch.tick()
+        self.assertEqual(self.activated, [])
+        self.assertTrue(watch.running)    # still waiting for the moment
+
+    def test_it_gives_up_rather_than_watching_for_ever(self):
+        watch = self.watching(4242, dikte_in_front=False)
+        with mock.patch.object(self.dikte.time, "monotonic",
+                               return_value=self.dikte.time.monotonic() + 60):
+            watch.tick()
+        self.assertFalse(watch.running)
+        self.assertEqual(self.activated, [])
+
+    def test_a_dictation_started_in_dikte_itself_watches_nothing(self):
+        """Settings is a window of ours, and the front is already where it
+        belongs."""
+        self.assertIsNone(self.watching(os.getpid(), dikte_in_front=True))
+
+    def test_a_second_recording_calls_off_the_watch_the_first_one_left(self):
+        """The older watch remembers where the older recording started, and by
+        now that is the wrong window to be pulling forward."""
+        dikte = self.bare()
+        first = self.watching(4242, dikte_in_front=False, on=dikte)
+        second = self.watching(1111, dikte_in_front=False, on=dikte)
+        self.assertFalse(first.running)
+        self.assertTrue(second.running)
+        second.tick()                      # and the survivor is the new one
+        self.patch_attr(self.mac_window_module(), "is_frontmost", lambda: True)
+        second.tick()
+        self.assertEqual(self.activated, [1111])
+
+    def test_a_recording_nobody_needs_watching_for_still_calls_off_the_old_one(self):
+        """Starting the next one from Dikte's own window is not a reason to
+        leave the last one's watch running."""
+        dikte = self.bare()
+        first = self.watching(4242, dikte_in_front=False, on=dikte)
+        self.watching(os.getpid(), dikte_in_front=False, on=dikte)
+        self.assertFalse(first.running)
+        self.assertIsNone(dikte._front_watch)
+
+    def test_nobody_in_front_is_nobody_to_go_back_to(self):
+        self.assertIsNone(self.watching(None, dikte_in_front=True))
+
+    def mac_window_module(self):
+        from dikte import mac_window
+        return mac_window
+
+
+class EveryRecordingProtectsTheFront(DikteTest):
+    """Three ways in, dictation, agent and meeting, and all three open the same
+    avfoundation capture, so all three take the front the same way. What is
+    checked here is the order: the front has to be noted before the microphone
+    is opened, and the watch armed after, or there is nothing to go back to.
+    """
+
+    def setUp(self):
+        super().setUp()
+        from dikte import app as dikte_module
+        self.dikte = dikte_module
+        self.order = []
+
+    def app(self, **attributes):
+        """A Dikte that records the order it does things in, and nothing else.
+
+        The methods under test are called unbound against it, so the stand-ins
+        go on the object rather than on the class.
+        """
+        dikte = mock.Mock(**attributes)
+        dikte._run_id = 0
+        dikte.conf = {"mic_target": "", "max_seconds": 60,
+                      "meeting_mic_target": "", "meeting_system_target": "",
+                      "meeting_max_seconds": 60}
+        dikte._the_front.side_effect = lambda: self.order.append("noted") or 4242
+        dikte._give_the_front_back.side_effect = (
+            lambda pid: self.order.append(f"watching {pid}"))
+        dikte._begin_recording = (
+            lambda owner: self.dikte.Dikte._begin_recording(dikte, owner))
+        dikte.recorder.start.side_effect = (
+            lambda *_a: self.order.append("microphone"))
+        dikte.meeting_recorder.start.side_effect = (
+            lambda *_a: self.order.append("microphone"))
+        return dikte
+
+    def test_a_dictation_notes_the_front_before_the_indicator_is_even_shown(self):
+        dikte = self.app(state=self.dikte.IDLE, recording=False)
+        dikte.overlay.show_recording.side_effect = (
+            lambda *_a: self.order.append("indicator"))
+        self.dikte.Dikte.start(dikte)
+        self.assertEqual(self.order,
+                         ["noted", "indicator", "microphone", "watching 4242"])
+
+    def test_the_agent_does_the_same(self):
+        dikte = self.app(ask_state=self.dikte.IDLE, recording=False)
+        self.dikte.Dikte.start_ask(dikte)
+        self.assertEqual(self.order, ["noted", "microphone", "watching 4242"])
+
+    def test_a_meeting_does_the_same(self):
+        """The one most worth protecting: the user is in a call."""
+        dikte = self.app(meeting_state=self.dikte.M_IDLE)
+        dikte.meeting_recorder.active = True
+        self.dikte.Dikte.start_meeting(dikte)
+        self.assertEqual(self.order, ["noted", "microphone", "watching 4242"])
+
+    def test_a_meeting_whose_microphone_never_opened_watches_nothing(self):
+        dikte = self.app(meeting_state=self.dikte.M_IDLE)
+        dikte.meeting_recorder.active = False
+        self.dikte.Dikte.start_meeting(dikte)
+        self.assertEqual(self.order, ["noted", "microphone"])
+
 class Overlay(DikteTest):
     def overlay(self, **kwargs):
         widget = overlay_module.Overlay(**kwargs)
@@ -658,7 +1003,9 @@ class MeetingSources(DikteTest):
                 only_these_tools(), \
                 mock.patch.object(settings_ui.SettingsWindow, "_load_models"), \
                 mock.patch.object(settings_ui.SettingsWindow,
-                                  "_load_transcribe_models"):
+                                  "_load_transcribe_models"), \
+                mock.patch.object(settings_ui.SettingsWindow,
+                                  "_load_codex_models"):
             window = settings_ui.SettingsWindow(cfg.Config())
         self.addCleanup(window.deleteLater)
         self.addCleanup(window.close)
@@ -687,6 +1034,9 @@ class LocalModels(DikteTest):
         # "nothing can transcribe" question from its real binary and model.
         self.patch_attr(ggml, "BIN_DIR", self.path("bin"))
         self.patch_attr(ggml, "MODELS_DIR", self.path("models"))
+        # And one with Codex on it would ask it for its model list.
+        self.enterContext(mock.patch.object(settings_ui.SettingsWindow,
+                                            "_load_codex_models"))
 
     def window(self, conf):
         window = settings_ui.SettingsWindow(conf)
@@ -723,12 +1073,21 @@ class LocalModels(DikteTest):
         # Qt's int is C++'s 32-bit one, and a 2.3 GB model is more than fits in
         # it: the count came out the far side negative, at "-1%".
         box = self.window(cfg.Config()).local_llm
-        box._downloading = True
-        box._report(1_048_576, 2_489_757_856)
+        box._report("model", 1_048_576, 2_489_757_856)
         _app.processEvents()
         self.assertIn("2.3 GB", box.status.text())
         self.assertNotIn("-", box.status.text())
 
+    def test_each_download_reports_into_its_own_label(self):
+        """The two can run at once; the tag, not a flag read later, says
+        which label the bytes belong to."""
+        box = self.window(cfg.Config()).local_llm
+        box._report("program", 10, 100)
+        box._report("model", 20, 100)
+        _app.processEvents()
+        self.assertIn("10", box.program_label.text())
+        self.assertIn("20", box.status.text())
+
     def test_a_long_model_name_is_not_cut_in_half(self):
         # The list under a combo box takes the box's width and elides what does
         # not fit, in the middle: "ggml-org/Qwen....7B-Base-GGUF".

tests/test_vad.py

@@ -128,6 +128,12 @@ class Hallucinations(DikteTest):
         self.assertFalse(vad.looks_like_hallucination("Bugün toplantı var.", 2.0))
         self.assertFalse(vad.looks_like_hallucination("Send it on Thursday.", 2.0))
 
+    def test_a_one_word_answer_is_believed(self):
+        # Whisper invents both over silence, but people dictate both as whole
+        # answers, and losing a real answer costs more than passing a fake one.
+        self.assertFalse(vad.looks_like_hallucination("You.", 1.5))
+        self.assertFalse(vad.looks_like_hallucination("Bye.", 1.5))
+
     def test_an_empty_transcript_counts_as_invented(self):
         self.assertTrue(vad.looks_like_hallucination("   ", 2.0))
         self.assertTrue(vad.looks_like_hallucination("...", 2.0))
@@ -138,8 +144,8 @@ class Hallucinations(DikteTest):
                 self.assertTrue(vad.looks_like_hallucination(text, 2.0))
 
     def test_the_boundary_is_the_max_duration(self):
-        self.assertTrue(vad.looks_like_hallucination("you", 6.0))
-        self.assertFalse(vad.looks_like_hallucination("you", 6.1))
+        self.assertTrue(vad.looks_like_hallucination("thanks for watching", 6.0))
+        self.assertFalse(vad.looks_like_hallucination("thanks for watching", 6.1))
 
 
 if __name__ == "__main__":

tests/test_worker.py

@@ -8,6 +8,7 @@ afterwards. A pull request that reorders any of it shows up here.
 import contextlib
 import io
 import os
+import threading
 import unittest
 from unittest import mock
 
@@ -30,9 +31,11 @@ class Chain(DikteTest):
 
     def run_chain(self, ask=False, paste_override=None, duration=2.0,
                   transcript="uh, book it for Thursday",
+                  transcribe_error=None,
                   cleaned="Book it for Thursday.",
                   cleanup_error=None, answer=("Booked.", ""), rms=None,
-                  clipboard=b"what was there before", paste_error=None):
+                  clipboard=b"what was there before", paste_error=None,
+                  focus=None):
         pipeline = worker.Pipeline(self.conf)
         done, failures, stages, cancels = [], [], [], []
         pipeline.finished.connect(lambda *args: done.append(args))
@@ -46,7 +49,10 @@ class Chain(DikteTest):
         # The chain reports its own failures on stderr, which a test run has no
         # use for.
         with contextlib.redirect_stderr(io.StringIO()), \
-                mock.patch.object(api, "transcribe", return_value=transcript) as tr, \
+                mock.patch.object(
+                    api, "transcribe",
+                    **({"side_effect": transcribe_error} if transcribe_error
+                       else {"return_value": transcript})) as tr, \
                 mock.patch.object(api, "cleanup", cleanup), \
                 mock.patch.object(assistant, "ask", return_value=answer) as ask_call, \
                 mock.patch.object(paste, "copy") as copy, \
@@ -60,7 +66,8 @@ class Chain(DikteTest):
                      "copy": copy, "copy_bytes": copy_bytes, "press": press,
                      "read_clipboard": read_clipboard}
             pipeline._work(self.wav, duration,
-                           self.rms if rms is None else rms, ask, paste_override)
+                           self.rms if rms is None else rms, ask, paste_override,
+                           focus)
         return {"done": done, "failures": failures, "stages": stages,
                 "cancelled": cancels, **calls}
 
@@ -72,7 +79,15 @@ class Chain(DikteTest):
         self.assertEqual(run["done"][0],
                          ("uh, book it for Thursday", "Book it for Thursday.", ""))
         run["copy"].assert_called_once_with("Book it for Thursday.")
-        run["press"].assert_called_once_with(self.conf["paste_shortcut"])
+        run["press"].assert_called_once_with(self.conf["paste_shortcut"],
+                                             focus=None)
+
+    def test_the_paste_is_told_where_the_dictation_started(self):
+        """Whoever was in front when the recording began is where the keys are
+        meant to go, and the press is the only part that can act on it."""
+        run = self.run_chain(focus=4242)
+        run["press"].assert_called_once_with(self.conf["paste_shortcut"],
+                                             focus=4242)
 
     def test_the_stages_are_named_as_they_happen(self):
         run = self.run_chain()
@@ -108,11 +123,57 @@ class Chain(DikteTest):
         run = self.run_chain()
         run["copy_bytes"].assert_not_called()
 
-    def test_the_clipboard_is_put_back_when_the_keypress_fails(self):
+    def test_a_failed_keypress_leaves_the_transcript_on_the_clipboard(self):
+        """The press failing is a warning, not a lost dictation: restoring the
+        old clipboard over the text would leave nothing to paste by hand."""
         self.conf["restore_clipboard"] = True
         run = self.run_chain(paste_error=paste.PasteError("not trusted"))
-        self.assertIn("not trusted", run["failures"][0])
-        run["copy_bytes"].assert_called_once_with(b"what was there before")
+        self.assertEqual(run["failures"], [])
+        raw, text, warning = run["done"][0]
+        self.assertIn("not trusted", warning)
+        run["copy_bytes"].assert_not_called()
+
+    def test_a_failed_keypress_still_reaches_the_history(self):
+        self.run_chain(paste_error=paste.PasteError("not trusted"))
+        rows = cfg.read_history()
+        self.assertEqual(len(rows), 1)
+        self.assertEqual(rows[0]["text"], "Book it for Thursday.")
+        # The row goes in before the paste is attempted, so the paste failing
+        # has to be written back into it: the record tells the whole truth.
+        self.assertIn("not trusted", rows[0]["cleanup_error"])
+
+    def test_a_failed_transcription_keeps_the_audio(self):
+        """Speech the user cannot repeat from memory must survive the failure."""
+        run = self.run_chain(transcribe_error=api.ApiError("server down"))
+        self.assertIn("server down", run["failures"][0])
+        self.assertIn("kept", run["failures"][0])
+        kept = list(cfg.RECORDINGS_DIR.glob("*.wav"))
+        self.assertEqual(len(kept), 1)
+        self.assertFalse(os.path.exists(self.wav))
+
+    def test_two_failures_in_one_second_keep_both_recordings(self):
+        self.run_chain(transcribe_error=api.ApiError("down"))
+        self.wav = make_wav(self.path("clip2.wav"), speech(2.0))
+        with mock.patch.object(worker.time, "strftime",
+                               return_value="20260820-120000"):
+            self.run_chain(transcribe_error=api.ApiError("down"))
+            self.wav = make_wav(self.path("clip3.wav"), speech(2.0))
+            self.run_chain(transcribe_error=api.ApiError("down"))
+        self.assertEqual(len(list(cfg.RECORDINGS_DIR.glob("*.wav"))), 3)
+
+    def test_the_history_row_says_whether_cleanup_actually_ran(self):
+        """The ask path cleans under its own setting; the record follows the
+        run, not the dictation gate."""
+        self.conf["cleanup_enabled"] = False
+        self.conf["assistant_cleanup"] = True
+        self.run_chain(ask=True)
+        row = cfg.read_history()[0]
+        self.assertNotEqual(row["cleanup_model"], "")
+        cfg.clear_history()
+        self.conf["cleanup_enabled"] = True
+        self.conf["assistant_cleanup"] = False
+        self.run_chain(ask=True)
+        self.assertEqual(cfg.read_history()[0]["cleanup_model"], "")
 
     def test_the_transcription_is_told_the_language_and_the_glossary(self):
         self.conf["language"] = "tr"
@@ -278,13 +339,41 @@ class Chain(DikteTest):
 
 
 class Busy(DikteTest):
-    def test_a_second_run_while_one_is_going_is_ignored(self):
+    def test_a_second_run_while_one_is_going_waits_its_turn(self):
+        """The microphone is free while a transcript is being cleaned up, so
+        the next dictation can already have been spoken by then. It has to run
+        once the first is done, in the order they were spoken, on one thread."""
         pipeline = worker.Pipeline(self.config())
-        pipeline._thread = mock.Mock(is_alive=lambda: True)
-        self.assertTrue(pipeline.busy)
-        with mock.patch.object(worker.threading, "Thread") as thread:
-            pipeline.run("/tmp/nope.wav", 1.0)
-        thread.assert_not_called()
+        order = []
+        started, gate = threading.Event(), threading.Event()
+
+        def work(wav_path, *_rest):
+            order.append(wav_path)
+            started.set()
+            if wav_path == "first.wav":
+                gate.wait(5)
+
+        with mock.patch.object(pipeline, "_work", side_effect=work):
+            pipeline.run("first.wav", 1.0)
+            self.assertTrue(started.wait(5))
+            pipeline.run("second.wav", 1.0)
+            # Held, not dropped and not running beside the first.
+            self.assertEqual(order, ["first.wav"])
+            gate.set()
+            pipeline._thread.join(5)
+        self.assertEqual(order, ["first.wav", "second.wav"])
+
+    def test_a_run_arriving_after_the_queue_drained(self):
+        """The worker thread ends with the queue; the next run brings one."""
+        pipeline = worker.Pipeline(self.config())
+        order = []
+        with mock.patch.object(pipeline, "_work",
+                               side_effect=lambda wav, *rest: order.append(wav)):
+            pipeline.run("first.wav", 1.0)
+            pipeline._thread.join(5)
+            pipeline.run("second.wav", 1.0)
+            pipeline._thread.join(5)
+        self.assertEqual(order, ["first.wav", "second.wav"])
 
     def test_the_chunk_length_matches_the_level_meter(self):
         """The silence thresholds are read in seconds, so the two must agree."""