python-voiceio 1.0.0__tar.gz → 1.2.0__tar.gz

{python_voiceio-1.0.0/python_voiceio.egg-info → python_voiceio-1.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-voiceio
-Version: 1.0.0
+Version: 1.2.0
 Summary: Voice dictation for Linux. Speak → text, locally, instantly.
 Author: Hugo Montenegro
 License-Expression: MIT
@@ -140,7 +140,7 @@ uv run voiceio setup      # also links voiceio into ~/.local/bin
 
 ## Your data
 
-Everything voiceio keeps is in three directories (they follow `XDG_CONFIG_HOME`, `XDG_STATE_HOME` and `XDG_DATA_HOME`). Files are created `0600`, directories `0700`. Nothing is uploaded.
+Everything voiceio keeps is in three directories (they follow `XDG_CONFIG_HOME`, `XDG_STATE_HOME` and `XDG_DATA_HOME`). Files are created `0600`, directories `0700`. Nothing is uploaded. Recordings, decode traces and window titles are **off unless you turn them on** — `voiceio setup` asks whether to keep recordings so voiceio can learn your voice. Logs record sizes and timings, never your words.
 
 | Path | Holds | Retention |
 |---|---|---|
@@ -150,14 +150,13 @@ Everything voiceio keeps is in three directories (they follow `XDG_CONFIG_HOME`,
 | `vocab_stats.json` | Per-term usage counts used to rank the vocabulary | Rebuilt as you dictate |
 | `corrections.json` | Find/replace rules (`post grass` → `PostgreSQL`) | — |
 | `flagged.txt` | Words flagged with the "correct that" voice command, for review | — |
-| `snapshots/` | Copies of corrections + vocabulary from before 1.0's cleanup (safe to delete) | Last 8 kept |
+| `snapshots/` | Copies of corrections + vocabulary from before 1.0 (safe to delete) | — |
 | `consent.json` | Whether you allowed cloud LLM calls (text only) | — |
 | `hints.json` | Which CLI tips you have already seen | — |
 | **`~/.local/state/voiceio/`** | *What you said* | |
-| `history.jsonl` | Every committed transcript: time, raw and final text, focused-window title | `[history] enabled`, `max_entries`, `max_age_days` (0 = keep forever) |
-| `recordings/*.wav` | The audio of each utterance, for `voiceio learn` | `[data] retain_audio`, `max_audio_mb` (default 4096), `min_free_gb` (default 5) |
-| `streaming_trace.jsonl`, `postcorrect_pairs.jsonl` | Per-pass decode traces and LLM before/after pairs | `[data] capture_intermediates`; each capped at 64 MB |
-| `corrections_audit.jsonl` | When each correction rule fired (shows which rules earn their keep) | — |
+| `history.jsonl` | Every committed transcript: time, raw and final text (and the focused-window title with `[data] capture_context`) | `[history] enabled`, `max_entries`, `max_age_days` (0 = keep forever) |
+| `recordings/*.wav` | The audio of each utterance, for `voiceio learn` — **opt-in** | `[data] retain_audio` (default off), `max_audio_mb` (default 4096), `min_free_gb` (default 5) |
+| `streaming_trace.jsonl`, `postcorrect_pairs.jsonl` | Per-pass decode text and timings, LLM before/after pairs — **opt-in**, for debugging | `[data] capture_intermediates` (default off); each capped at 64 MB |
 | `voiceio.log`, `worker.log`, `ibus-engine-stderr.log` | Logs | `voiceio.log` rotates at 2 MB × 3 |
 | **`~/.local/share/voiceio/learn/`** | *Personalization corpus* (only exists once you run `voiceio learn`) | Never pruned; delete freely |
 | `labels.jsonl` | Reference transcripts from the teacher model or your review | — |
@@ -191,6 +190,7 @@ voiceio learn eval       # score the current model on your held-out clips: WER a
 voiceio learn train      # LoRA fine-tune (needs the `train` extra; decoder-only on CPU, --full on a GPU)
 voiceio learn promote <run>   # dictate with the fine-tune; `voiceio learn rollback` undoes it
 voiceio learn maintain   # later: label what's new, and once 50+ clips piled up, retrain and compare
+voiceio learn schedule on   # …or let it run `maintain` every night, and notify you when a model wins
 ```
 
 Labeling, training and offline eval run in an idle-priority cgroup, so they never slow dictation down. Everything stays on your machine unless you choose `train --remote`.
@@ -292,7 +292,7 @@ There is a best-effort pynput path for both (hotkeys and typing, no live preview
 | Hotkey does nothing on Wayland | `voiceio doctor --fix` installs the keyboard udev rule, or bind `voiceio toggle` in your compositor |
 | `sounddevice is not installed` | You installed without the desktop extra: `pipx install --force 'python-voiceio[desktop]'` |
 | evdev fails to build | Install a C toolchain and Python headers (see Quickstart), then reinstall |
-| Too slow | `voiceio doctor` shows your real stop→text; turn off cloud postcorrect, or `voiceio models use small` (or a GPU with `large-v3-turbo`) |
+| Too slow | `voiceio doctor` shows your real stop→text. Turn off cloud postcorrect; use `voiceio models use small` (or a GPU with `large-v3-turbo`); or narrow the beam of the passes you keep: on the author's 44 clips `[model] beam_size = 2` matched the default 5 (WER 4.9% vs 5.4%, vocabulary 28/32 vs 29/32) while decoding faster; greedy (`1`) lost vocabulary (26/32). Measure on yours: `voiceio learn eval --beam 2` |
 | Words come out wrong in a pattern | `voiceio learn eval --pipeline` and `voiceio corrections audit` show which stage or rule did it |
 | Start over | `voiceio uninstall`, then `voiceio setup` |
 
@@ -300,8 +300,8 @@ Logs: `journalctl --user -u voiceio` or `~/.local/state/voiceio/voiceio.log`.
 
 ## Roadmap
 
-- **A landing page** — voiceio as local, open-source, private dictation that learns from you.
-- **Continuous learning** — `learn maintain` on a timer: label new dictation while idle, retrain when enough piles up, and promote only what wins on your held-out clips.
+- **A landing page** — voiceio as local, open-source, private dictation that learns from you. A draft is in [`site/index.html`](site/index.html), ready for GitHub Pages.
+- **Hands-off promotion** — `learn schedule` already retrains nightly and tells you when a model wins; next is switching to it automatically, with one-command rollback.
 - **Unbounded vocabulary** — move to a decoder whose biasing has no token cap (see [model choice](CONTRIBUTING.md#model-choice-why-whisper-and-what-would-change-it)).
 - **One model everywhere** — sync your fine-tune to your phone server and laptops, peer to peer.
 

{python_voiceio-1.0.0 → python_voiceio-1.2.0}/README.md

@@ -87,7 +87,7 @@ uv run voiceio setup      # also links voiceio into ~/.local/bin
 
 ## Your data
 
-Everything voiceio keeps is in three directories (they follow `XDG_CONFIG_HOME`, `XDG_STATE_HOME` and `XDG_DATA_HOME`). Files are created `0600`, directories `0700`. Nothing is uploaded.
+Everything voiceio keeps is in three directories (they follow `XDG_CONFIG_HOME`, `XDG_STATE_HOME` and `XDG_DATA_HOME`). Files are created `0600`, directories `0700`. Nothing is uploaded. Recordings, decode traces and window titles are **off unless you turn them on** — `voiceio setup` asks whether to keep recordings so voiceio can learn your voice. Logs record sizes and timings, never your words.
 
 | Path | Holds | Retention |
 |---|---|---|
@@ -97,14 +97,13 @@ Everything voiceio keeps is in three directories (they follow `XDG_CONFIG_HOME`,
 | `vocab_stats.json` | Per-term usage counts used to rank the vocabulary | Rebuilt as you dictate |
 | `corrections.json` | Find/replace rules (`post grass` → `PostgreSQL`) | — |
 | `flagged.txt` | Words flagged with the "correct that" voice command, for review | — |
-| `snapshots/` | Copies of corrections + vocabulary from before 1.0's cleanup (safe to delete) | Last 8 kept |
+| `snapshots/` | Copies of corrections + vocabulary from before 1.0 (safe to delete) | — |
 | `consent.json` | Whether you allowed cloud LLM calls (text only) | — |
 | `hints.json` | Which CLI tips you have already seen | — |
 | **`~/.local/state/voiceio/`** | *What you said* | |
-| `history.jsonl` | Every committed transcript: time, raw and final text, focused-window title | `[history] enabled`, `max_entries`, `max_age_days` (0 = keep forever) |
-| `recordings/*.wav` | The audio of each utterance, for `voiceio learn` | `[data] retain_audio`, `max_audio_mb` (default 4096), `min_free_gb` (default 5) |
-| `streaming_trace.jsonl`, `postcorrect_pairs.jsonl` | Per-pass decode traces and LLM before/after pairs | `[data] capture_intermediates`; each capped at 64 MB |
-| `corrections_audit.jsonl` | When each correction rule fired (shows which rules earn their keep) | — |
+| `history.jsonl` | Every committed transcript: time, raw and final text (and the focused-window title with `[data] capture_context`) | `[history] enabled`, `max_entries`, `max_age_days` (0 = keep forever) |
+| `recordings/*.wav` | The audio of each utterance, for `voiceio learn` — **opt-in** | `[data] retain_audio` (default off), `max_audio_mb` (default 4096), `min_free_gb` (default 5) |
+| `streaming_trace.jsonl`, `postcorrect_pairs.jsonl` | Per-pass decode text and timings, LLM before/after pairs — **opt-in**, for debugging | `[data] capture_intermediates` (default off); each capped at 64 MB |
 | `voiceio.log`, `worker.log`, `ibus-engine-stderr.log` | Logs | `voiceio.log` rotates at 2 MB × 3 |
 | **`~/.local/share/voiceio/learn/`** | *Personalization corpus* (only exists once you run `voiceio learn`) | Never pruned; delete freely |
 | `labels.jsonl` | Reference transcripts from the teacher model or your review | — |
@@ -138,6 +137,7 @@ voiceio learn eval       # score the current model on your held-out clips: WER a
 voiceio learn train      # LoRA fine-tune (needs the `train` extra; decoder-only on CPU, --full on a GPU)
 voiceio learn promote <run>   # dictate with the fine-tune; `voiceio learn rollback` undoes it
 voiceio learn maintain   # later: label what's new, and once 50+ clips piled up, retrain and compare
+voiceio learn schedule on   # …or let it run `maintain` every night, and notify you when a model wins
 ```
 
 Labeling, training and offline eval run in an idle-priority cgroup, so they never slow dictation down. Everything stays on your machine unless you choose `train --remote`.
@@ -239,7 +239,7 @@ There is a best-effort pynput path for both (hotkeys and typing, no live preview
 | Hotkey does nothing on Wayland | `voiceio doctor --fix` installs the keyboard udev rule, or bind `voiceio toggle` in your compositor |
 | `sounddevice is not installed` | You installed without the desktop extra: `pipx install --force 'python-voiceio[desktop]'` |
 | evdev fails to build | Install a C toolchain and Python headers (see Quickstart), then reinstall |
-| Too slow | `voiceio doctor` shows your real stop→text; turn off cloud postcorrect, or `voiceio models use small` (or a GPU with `large-v3-turbo`) |
+| Too slow | `voiceio doctor` shows your real stop→text. Turn off cloud postcorrect; use `voiceio models use small` (or a GPU with `large-v3-turbo`); or narrow the beam of the passes you keep: on the author's 44 clips `[model] beam_size = 2` matched the default 5 (WER 4.9% vs 5.4%, vocabulary 28/32 vs 29/32) while decoding faster; greedy (`1`) lost vocabulary (26/32). Measure on yours: `voiceio learn eval --beam 2` |
 | Words come out wrong in a pattern | `voiceio learn eval --pipeline` and `voiceio corrections audit` show which stage or rule did it |
 | Start over | `voiceio uninstall`, then `voiceio setup` |
 
@@ -247,8 +247,8 @@ Logs: `journalctl --user -u voiceio` or `~/.local/state/voiceio/voiceio.log`.
 
 ## Roadmap
 
-- **A landing page** — voiceio as local, open-source, private dictation that learns from you.
-- **Continuous learning** — `learn maintain` on a timer: label new dictation while idle, retrain when enough piles up, and promote only what wins on your held-out clips.
+- **A landing page** — voiceio as local, open-source, private dictation that learns from you. A draft is in [`site/index.html`](site/index.html), ready for GitHub Pages.
+- **Hands-off promotion** — `learn schedule` already retrains nightly and tells you when a model wins; next is switching to it automatically, with one-command rollback.
 - **Unbounded vocabulary** — move to a decoder whose biasing has no token cap (see [model choice](CONTRIBUTING.md#model-choice-why-whisper-and-what-would-change-it)).
 - **One model everywhere** — sync your fine-tune to your phone server and laptops, peer to peer.
 

{python_voiceio-1.0.0 → python_voiceio-1.2.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "python-voiceio"
-version = "1.0.0"
+version = "1.2.0"
 description = "Voice dictation for Linux. Speak → text, locally, instantly."
 readme = "README.md"
 license = "MIT"

{python_voiceio-1.0.0 → python_voiceio-1.2.0/python_voiceio.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-voiceio
-Version: 1.0.0
+Version: 1.2.0
 Summary: Voice dictation for Linux. Speak → text, locally, instantly.
 Author: Hugo Montenegro
 License-Expression: MIT
@@ -140,7 +140,7 @@ uv run voiceio setup      # also links voiceio into ~/.local/bin
 
 ## Your data
 
-Everything voiceio keeps is in three directories (they follow `XDG_CONFIG_HOME`, `XDG_STATE_HOME` and `XDG_DATA_HOME`). Files are created `0600`, directories `0700`. Nothing is uploaded.
+Everything voiceio keeps is in three directories (they follow `XDG_CONFIG_HOME`, `XDG_STATE_HOME` and `XDG_DATA_HOME`). Files are created `0600`, directories `0700`. Nothing is uploaded. Recordings, decode traces and window titles are **off unless you turn them on** — `voiceio setup` asks whether to keep recordings so voiceio can learn your voice. Logs record sizes and timings, never your words.
 
 | Path | Holds | Retention |
 |---|---|---|
@@ -150,14 +150,13 @@ Everything voiceio keeps is in three directories (they follow `XDG_CONFIG_HOME`,
 | `vocab_stats.json` | Per-term usage counts used to rank the vocabulary | Rebuilt as you dictate |
 | `corrections.json` | Find/replace rules (`post grass` → `PostgreSQL`) | — |
 | `flagged.txt` | Words flagged with the "correct that" voice command, for review | — |
-| `snapshots/` | Copies of corrections + vocabulary from before 1.0's cleanup (safe to delete) | Last 8 kept |
+| `snapshots/` | Copies of corrections + vocabulary from before 1.0 (safe to delete) | — |
 | `consent.json` | Whether you allowed cloud LLM calls (text only) | — |
 | `hints.json` | Which CLI tips you have already seen | — |
 | **`~/.local/state/voiceio/`** | *What you said* | |
-| `history.jsonl` | Every committed transcript: time, raw and final text, focused-window title | `[history] enabled`, `max_entries`, `max_age_days` (0 = keep forever) |
-| `recordings/*.wav` | The audio of each utterance, for `voiceio learn` | `[data] retain_audio`, `max_audio_mb` (default 4096), `min_free_gb` (default 5) |
-| `streaming_trace.jsonl`, `postcorrect_pairs.jsonl` | Per-pass decode traces and LLM before/after pairs | `[data] capture_intermediates`; each capped at 64 MB |
-| `corrections_audit.jsonl` | When each correction rule fired (shows which rules earn their keep) | — |
+| `history.jsonl` | Every committed transcript: time, raw and final text (and the focused-window title with `[data] capture_context`) | `[history] enabled`, `max_entries`, `max_age_days` (0 = keep forever) |
+| `recordings/*.wav` | The audio of each utterance, for `voiceio learn` — **opt-in** | `[data] retain_audio` (default off), `max_audio_mb` (default 4096), `min_free_gb` (default 5) |
+| `streaming_trace.jsonl`, `postcorrect_pairs.jsonl` | Per-pass decode text and timings, LLM before/after pairs — **opt-in**, for debugging | `[data] capture_intermediates` (default off); each capped at 64 MB |
 | `voiceio.log`, `worker.log`, `ibus-engine-stderr.log` | Logs | `voiceio.log` rotates at 2 MB × 3 |
 | **`~/.local/share/voiceio/learn/`** | *Personalization corpus* (only exists once you run `voiceio learn`) | Never pruned; delete freely |
 | `labels.jsonl` | Reference transcripts from the teacher model or your review | — |
@@ -191,6 +190,7 @@ voiceio learn eval       # score the current model on your held-out clips: WER a
 voiceio learn train      # LoRA fine-tune (needs the `train` extra; decoder-only on CPU, --full on a GPU)
 voiceio learn promote <run>   # dictate with the fine-tune; `voiceio learn rollback` undoes it
 voiceio learn maintain   # later: label what's new, and once 50+ clips piled up, retrain and compare
+voiceio learn schedule on   # …or let it run `maintain` every night, and notify you when a model wins
 ```
 
 Labeling, training and offline eval run in an idle-priority cgroup, so they never slow dictation down. Everything stays on your machine unless you choose `train --remote`.
@@ -292,7 +292,7 @@ There is a best-effort pynput path for both (hotkeys and typing, no live preview
 | Hotkey does nothing on Wayland | `voiceio doctor --fix` installs the keyboard udev rule, or bind `voiceio toggle` in your compositor |
 | `sounddevice is not installed` | You installed without the desktop extra: `pipx install --force 'python-voiceio[desktop]'` |
 | evdev fails to build | Install a C toolchain and Python headers (see Quickstart), then reinstall |
-| Too slow | `voiceio doctor` shows your real stop→text; turn off cloud postcorrect, or `voiceio models use small` (or a GPU with `large-v3-turbo`) |
+| Too slow | `voiceio doctor` shows your real stop→text. Turn off cloud postcorrect; use `voiceio models use small` (or a GPU with `large-v3-turbo`); or narrow the beam of the passes you keep: on the author's 44 clips `[model] beam_size = 2` matched the default 5 (WER 4.9% vs 5.4%, vocabulary 28/32 vs 29/32) while decoding faster; greedy (`1`) lost vocabulary (26/32). Measure on yours: `voiceio learn eval --beam 2` |
 | Words come out wrong in a pattern | `voiceio learn eval --pipeline` and `voiceio corrections audit` show which stage or rule did it |
 | Start over | `voiceio uninstall`, then `voiceio setup` |
 
@@ -300,8 +300,8 @@ Logs: `journalctl --user -u voiceio` or `~/.local/state/voiceio/voiceio.log`.
 
 ## Roadmap
 
-- **A landing page** — voiceio as local, open-source, private dictation that learns from you.
-- **Continuous learning** — `learn maintain` on a timer: label new dictation while idle, retrain when enough piles up, and promote only what wins on your held-out clips.
+- **A landing page** — voiceio as local, open-source, private dictation that learns from you. A draft is in [`site/index.html`](site/index.html), ready for GitHub Pages.
+- **Hands-off promotion** — `learn schedule` already retrains nightly and tells you when a model wins; next is switching to it automatically, with one-command rollback.
 - **Unbounded vocabulary** — move to a decoder whose biasing has no token cap (see [model choice](CONTRIBUTING.md#model-choice-why-whisper-and-what-would-change-it)).
 - **One model everywhere** — sync your fine-tune to your phone server and laptops, peer to peer.
 

{python_voiceio-1.0.0 → python_voiceio-1.2.0}/python_voiceio.egg-info/SOURCES.txt

@@ -76,6 +76,7 @@ voiceio/config.py
 voiceio/consent.py
 voiceio/corrections.py
 voiceio/demo.py
+voiceio/display.py
 voiceio/feedback.py
 voiceio/health.py
 voiceio/hints.py

{python_voiceio-1.0.0 → python_voiceio-1.2.0}/tests/test_app_wiring.py

@@ -13,18 +13,18 @@ import pytest
 from voiceio.config import Config
 
 
-def _make_vio(mock_transcriber=None):
+def _make_vio(mock_decoder=None):
     """Create a VoiceIO instance with mocked backends."""
     mock_hotkey = MagicMock()
     mock_hotkey.name = "socket"
     mock_typer = MagicMock()
     mock_typer.name = "clipboard"
-    if mock_transcriber is None:
-        mock_transcriber = FakeDecoder()
+    if mock_decoder is None:
+        mock_decoder = FakeDecoder()
 
     with patch("voiceio.app.hotkey_chain.select", return_value=mock_hotkey), \
          patch("voiceio.app.typer_chain.select", return_value=mock_typer), \
-         patch("voiceio.app.build_decoder", return_value=mock_transcriber), \
+         patch("voiceio.app.build_decoder", return_value=mock_decoder), \
          patch("voiceio.app.plat.detect") as mock_detect:
         mock_detect.return_value = MagicMock(display_server="wayland", desktop="gnome")
 
@@ -39,7 +39,7 @@ def _make_vio(mock_transcriber=None):
         mock_stream.stopped = False
         vio.recorder._stream = mock_stream  # skip real audio
         vio.recorder._last_callback_time = time.monotonic()  # healthy heartbeat
-        return vio, mock_typer, mock_transcriber
+        return vio, mock_typer, mock_decoder
 
 
 def _feed_audio(vio, chunks=20):

{python_voiceio-1.0.0 → python_voiceio-1.2.0}/tests/test_concurrency_lockdown.py

@@ -98,7 +98,7 @@ class TestGuardedSwaps:
         from voiceio.app import _State
         vio = _make_app()
         vio._state = _State.RECORDING
-        vio.transcriber.healthy.return_value = True
+        vio.decoder.healthy.return_value = True
         vio.recorder.stream_health = MagicMock(return_value=(True, ""))
         vio.cfg.tray.enabled = False
         vio._typers.health_upkeep = MagicMock()
@@ -153,7 +153,7 @@ class TestIBusGoodCitizen:
         from voiceio.app import _State
         vio = _make_app(typer_name="ibus")
         vio._state = _State.IDLE
-        vio.transcriber.healthy.return_value = True
+        vio.decoder.healthy.return_value = True
         vio.recorder.stream_health = MagicMock(return_value=(True, ""))
         vio.cfg.tray.enabled = False
         vio._typers.health_upkeep = MagicMock()

{python_voiceio-1.0.0 → python_voiceio-1.2.0}/tests/test_engine.py

@@ -67,6 +67,7 @@ def test_commit_strips_prefix_learns_and_keeps_history():
 
     from voiceio import config
     eng, _ = _engine(voice_input_prefix="[voice]")
+    eng.cfg.data.retain_audio = True
     stored = eng.commit("[voice] we deployed the Kalshi bot", raw="we deployed the kalshi bot",
                         duration=2.0, audio=np.zeros(16000, dtype=np.float32))
     assert stored == "we deployed the Kalshi bot"

{python_voiceio-1.0.0 → python_voiceio-1.2.0}/tests/test_health.py

@@ -105,3 +105,21 @@ def test_usage_decoder_counts_incidents_on_the_current_release(tmp_path, monkeyp
     assert ok
     assert finding.startswith(f"decoder: 0 hangs, 1 crashes in the last 7 days on {__version__}")
     assert f"(1 more before {__version__})" in finding
+
+
+def test_learning_status_and_tips(monkeypatch):
+    from voiceio import config, service
+    from voiceio.health import _accuracy_tips, _learning_status
+    from voiceio.learn.store import learn_dir
+    monkeypatch.setattr(service, "learn_timer_enabled", lambda: False)
+    cfg = config.Config()
+    st = _learning_status(cfg)
+    assert st["Voice model"] == "stock (small)" and st["Recordings"].startswith("off")
+    assert any("retain_audio true" in t for t in _accuracy_tips(st))
+
+    cfg.data.retain_audio = True
+    cfg.model.name = str(learn_dir() / "models" / "v0001-small-x" / "ct2")
+    st = _learning_status(cfg)
+    assert st["Voice model"] == "fine-tuned on you (v0001-small-x)"
+    assert st["Recordings"].startswith("on (") and st["Nightly learning"] == "off"
+    assert any("learn schedule on" in t for t in _accuracy_tips(st))

{python_voiceio-1.0.0 → python_voiceio-1.2.0}/tests/test_learn_pipeline.py

@@ -230,11 +230,27 @@ def test_maintain_waits_for_enough_new_clips(monkeypatch, capsys):
     store.add_label("new.wav", [Segment(0, 3, "a new one")], "teacher:fake")
 
     monkeypatch.setattr(cli, "cmd_label", lambda args: None)
+    monkeypatch.setattr(train, "missing_train_deps", lambda: [])
     monkeypatch.setattr(cli, "_train", lambda *a, **k: pytest.fail("must not retrain"))
     cli.cmd_maintain(argparse.Namespace(min_new=5, base="small", epochs=1))
     assert "1 new labeled clips since v0001" in capsys.readouterr().out
 
 
+def test_maintain_without_train_extra_labels_and_exits_cleanly(monkeypatch, capsys):
+    """A nightly timer must not fail every night, or spend an eval, over an
+    optional install."""
+    import argparse
+
+    from voiceio.cli import learn as cli
+    labeled = []
+    monkeypatch.setattr(cli, "cmd_label", lambda args: labeled.append(True))
+    monkeypatch.setattr(train, "missing_train_deps", lambda: ["torch", "peft"])
+    monkeypatch.setattr(store, "build_dataset", lambda: pytest.fail("must not build"))
+    monkeypatch.setattr(cli, "_run_eval", lambda *a, **k: pytest.fail("must not eval"))
+    cli.cmd_maintain(argparse.Namespace(min_new=5, base="small", epochs=1))
+    assert labeled and "python-voiceio[train]" in capsys.readouterr().out
+
+
 def test_dev_mix_is_stable_per_row():
     assert train._row_draws("a.wav") == train._row_draws("a.wav")
     assert train._row_draws("a.wav") != train._row_draws("b.wav")

{python_voiceio-1.0.0 → python_voiceio-1.2.0}/tests/test_postcorrect.py

@@ -17,6 +17,7 @@ def _cfg(*, enabled=True, api_key="test-key", min_words=4, model="",
     )
     cfg.output.remove_disfluencies = remove_disfluencies
     cfg.cloud.api_key = api_key
+    cfg.data.capture_intermediates = True   # most tests here check the pairs log
     return cfg
 
 
@@ -364,3 +365,14 @@ def test_guard_rejects_meaning_changes(original, corrected, reason):
 def test_guard_allows_filler_repeats_and_false_starts(original, corrected):
     pc = PostCorrector(_cfg(remove_disfluencies=True))
     assert pc._guard(original, corrected) == (True, "")
+
+
+def test_nothing_recorded_unless_opted_in(tmp_path, monkeypatch):
+    from voiceio import config as cfgmod
+    pairs = tmp_path / "pairs.jsonl"
+    monkeypatch.setattr(cfgmod, "POSTCORRECT_PAIRS_PATH", pairs)
+    cfg = _cfg()
+    cfg.data.capture_intermediates = False
+    with patch("voiceio.llm_api.chat", return_value="Crisp chat is the tool"):
+        PostCorrector(cfg).correct("Chris chat is the tool")
+    assert not pairs.exists()

{python_voiceio-1.0.0 → python_voiceio-1.2.0}/tests/test_postprocess.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 from voiceio.config import Config
 from voiceio.core.pipeline import Pipeline
 from voiceio.postprocess import cleanup, strip_disfluencies
-from voiceio.streaming import _word_match_len
+from voiceio.display import _word_match_len
 
 
 def _run(text: str, *, final: bool = False, **output) -> tuple[str, bool]:

{python_voiceio-1.0.0 → python_voiceio-1.2.0}/tests/test_retention.py

@@ -12,7 +12,7 @@ from voiceio.config import DataConfig
 
 class TestSaveAudio:
     def test_saves_wav(self, tmp_path):
-        cfg = DataConfig()
+        cfg = DataConfig(retain_audio=True)
         audio = np.full(16000, 0.1, dtype=np.float32)
         with patch.object(retention, "RECORDINGS_DIR", tmp_path):
             name = retention.save_audio(audio, ts=1700000000.5, cfg=cfg)
@@ -29,13 +29,13 @@ class TestSaveAudio:
             assert retention.save_audio(audio, ts=0, cfg=cfg) is None
 
     def test_none_and_empty_audio(self, tmp_path):
-        cfg = DataConfig()
+        cfg = DataConfig(retain_audio=True)
         with patch.object(retention, "RECORDINGS_DIR", tmp_path):
             assert retention.save_audio(None, ts=0, cfg=cfg) is None
             assert retention.save_audio(np.zeros(0, dtype=np.float32), ts=0, cfg=cfg) is None
 
     def test_roundtrip_amplitude(self, tmp_path):
-        cfg = DataConfig()
+        cfg = DataConfig(retain_audio=True)
         audio = np.full(1600, 0.5, dtype=np.float32)
         with patch.object(retention, "RECORDINGS_DIR", tmp_path):
             name = retention.save_audio(audio, ts=1700000001.0, cfg=cfg)
@@ -47,7 +47,7 @@ class TestSaveAudio:
 class TestPrune:
     def test_prunes_oldest_over_cap(self, tmp_path):
         import os
-        cfg = DataConfig(max_audio_mb=1)
+        cfg = DataConfig(retain_audio=True, max_audio_mb=1)
         # Three ~600KB files: total ~1.8MB > 1MB cap → oldest goes
         for i, name in enumerate(["a.wav", "b.wav", "c.wav"]):
             p = tmp_path / name
@@ -60,7 +60,7 @@ class TestPrune:
         assert "c.wav" in remaining
 
     def test_noop_under_cap(self, tmp_path):
-        cfg = DataConfig(max_audio_mb=100)
+        cfg = DataConfig(retain_audio=True, max_audio_mb=100)
         (tmp_path / "a.wav").write_bytes(b"\0" * 1000)
         with patch.object(retention, "RECORDINGS_DIR", tmp_path):
             retention.prune(cfg)
@@ -122,7 +122,7 @@ class TestSaveTrace:
         from voiceio import retention
         from voiceio.config import DataConfig
         monkeypatch.setattr(retention, "TRACES_PATH", tmp_path / "trace.jsonl")
-        retention.save_trace(DataConfig(), {"ts": 1, "passes": []})
+        retention.save_trace(DataConfig(capture_intermediates=True), {"ts": 1, "passes": []})
         assert (tmp_path / "trace.jsonl").exists()
 
     def test_skips_when_disabled(self, tmp_path, monkeypatch):
@@ -140,7 +140,7 @@ class TestDiskFloor:
         from voiceio.config import DataConfig
         monkeypatch.setattr(retention, "_free_gb", lambda p: 1.0)
         name = retention.save_audio(
-            np.zeros(1600, dtype=np.float32), 0.0, DataConfig(min_free_gb=5.0),
+            np.zeros(1600, dtype=np.float32), 0.0, DataConfig(retain_audio=True, min_free_gb=5.0),
         )
         assert name is None
 
@@ -151,7 +151,7 @@ class TestDiskFloor:
         monkeypatch.setattr(retention, "RECORDINGS_DIR", tmp_path)
         monkeypatch.setattr(retention, "_free_gb", lambda p: 100.0)
         name = retention.save_audio(
-            np.zeros(1600, dtype=np.float32), 0.0, DataConfig(min_free_gb=5.0),
+            np.zeros(1600, dtype=np.float32), 0.0, DataConfig(retain_audio=True, min_free_gb=5.0),
         )
         assert name is not None and (tmp_path / name).exists()
 

{python_voiceio-1.0.0 → python_voiceio-1.2.0}/tests/test_robustness.py

@@ -17,18 +17,18 @@ from voiceio.config import Config
 # Helpers (same pattern as test_app_wiring.py)
 # ---------------------------------------------------------------------------
 
-def _make_vio(mock_transcriber=None):
+def _make_vio(mock_decoder=None):
     """Create a VoiceIO instance with mocked backends."""
     mock_hotkey = MagicMock()
     mock_hotkey.name = "socket"
     mock_typer = MagicMock()
     mock_typer.name = "clipboard"
-    if mock_transcriber is None:
-        mock_transcriber = FakeDecoder()
+    if mock_decoder is None:
+        mock_decoder = FakeDecoder()
 
     with patch("voiceio.app.hotkey_chain.select", return_value=mock_hotkey), \
          patch("voiceio.app.typer_chain.select", return_value=mock_typer), \
-         patch("voiceio.app.build_decoder", return_value=mock_transcriber), \
+         patch("voiceio.app.build_decoder", return_value=mock_decoder), \
          patch("voiceio.app.plat.detect") as mock_detect:
         mock_detect.return_value = MagicMock(display_server="wayland", desktop="gnome")
 
@@ -40,7 +40,7 @@ def _make_vio(mock_transcriber=None):
         mock_stream.stopped = False
         vio.recorder._stream = mock_stream
         vio.recorder._last_callback_time = time.monotonic()
-        return vio, mock_typer, mock_transcriber
+        return vio, mock_typer, mock_decoder
 
 
 def _feed_audio(vio, chunks=20):
@@ -421,7 +421,7 @@ class TestHealthWatchdogAudioBackoff:
         vio, _, _ = _make_vio()
         vio.recorder.stream_health = MagicMock(return_value=(False, "stream not active"))
         vio.recorder.reopen_stream = MagicMock()
-        vio.transcriber.healthy = MagicMock(return_value=True)
+        vio.decoder.healthy = MagicMock(return_value=True)
 
         with patch("voiceio.watchdog.tray"):
             vio._watchdog.check()
@@ -434,7 +434,7 @@ class TestHealthWatchdogAudioBackoff:
         vio._watchdog._next_stream_retry = 0  # allow retry
         vio.recorder.stream_health = MagicMock(return_value=(False, "stream not active"))
         vio.recorder.reopen_stream = MagicMock()  # succeeds (no exception)
-        vio.transcriber.healthy = MagicMock(return_value=True)
+        vio.decoder.healthy = MagicMock(return_value=True)
 
         with patch("voiceio.watchdog.tray"):
             vio._watchdog.check()
@@ -448,7 +448,7 @@ class TestHealthWatchdogAudioBackoff:
         vio._watchdog._next_stream_retry = 0
         vio.recorder.stream_health = MagicMock(return_value=(False, "stream not active"))
         vio.recorder.reopen_stream = MagicMock(side_effect=OSError("device gone"))
-        vio.transcriber.healthy = MagicMock(return_value=True)
+        vio.decoder.healthy = MagicMock(return_value=True)
 
         with patch("voiceio.watchdog.tray"):
             vio._watchdog.check()
@@ -463,7 +463,7 @@ class TestHealthWatchdogAudioBackoff:
         vio._watchdog._next_stream_retry = time.monotonic() + 999  # far future
         vio.recorder.stream_health = MagicMock(return_value=(False, "stream not active"))
         vio.recorder.reopen_stream = MagicMock()
-        vio.transcriber.healthy = MagicMock(return_value=True)
+        vio.decoder.healthy = MagicMock(return_value=True)
 
         with patch("voiceio.watchdog.tray"):
             vio._watchdog.check()
@@ -476,7 +476,7 @@ class TestHealthWatchdogAudioBackoff:
         vio._watchdog._stream_fail_count = 3
         vio._watchdog._next_stream_retry = time.monotonic() + 100
         vio.recorder.stream_health = MagicMock(return_value=(True, ""))
-        vio.transcriber.healthy = MagicMock(return_value=True)
+        vio.decoder.healthy = MagicMock(return_value=True)
 
         with patch("voiceio.watchdog.tray"):
             vio._watchdog.check()
@@ -489,7 +489,7 @@ class TestHealthWatchdogAudioBackoff:
         vio, _, _ = _make_vio()
         vio.recorder.stream_health = MagicMock(return_value=(False, "stream not active"))
         vio.recorder.reopen_stream = MagicMock(side_effect=OSError("device gone"))
-        vio.transcriber.healthy = MagicMock(return_value=True)
+        vio.decoder.healthy = MagicMock(return_value=True)
 
         delays = []
         with patch("voiceio.watchdog.tray"):
@@ -551,7 +551,7 @@ class TestPreFlightStreamCheck:
 class TestStreamingWorkerExceptionHandling:
     def test_worker_catches_exception_and_continues(self):
         """Worker loop should catch exceptions in _transcribe_and_apply and keep running."""
-        mock_transcriber = FakeDecoder()
+        mock_decoder = FakeDecoder()
         mock_typer = MagicMock()
         mock_typer.name = "clipboard"
         mock_recorder = MagicMock()
@@ -561,7 +561,7 @@ class TestStreamingWorkerExceptionHandling:
         from voiceio.streaming import StreamingSession
 
         session = StreamingSession(
-            mock_transcriber, mock_typer, mock_recorder,
+            mock_decoder, mock_typer, mock_recorder,
             generation=0,
         )
 
@@ -593,8 +593,8 @@ class TestStreamingWorkerExceptionHandling:
 
     def test_final_transcription_exception_does_not_crash(self):
         """Exception during final transcription should be caught."""
-        mock_transcriber = FakeDecoder()
-        mock_transcriber.transcribe.side_effect = RuntimeError("model crash")
+        mock_decoder = FakeDecoder()
+        mock_decoder.transcribe.side_effect = RuntimeError("model crash")
         mock_typer = MagicMock()
         mock_typer.name = "clipboard"
         mock_recorder = MagicMock()
@@ -604,7 +604,7 @@ class TestStreamingWorkerExceptionHandling:
         from voiceio.streaming import StreamingSession
 
         session = StreamingSession(
-            mock_transcriber, mock_typer, mock_recorder,
+            mock_decoder, mock_typer, mock_recorder,
             generation=0,
         )
 
@@ -629,7 +629,7 @@ class TestTrayWatchdog:
         vio, _, _ = _make_vio()
         vio.cfg.tray.enabled = True
         vio.recorder.stream_health = MagicMock(return_value=(True, ""))
-        vio.transcriber.healthy = MagicMock(return_value=True)
+        vio.decoder.healthy = MagicMock(return_value=True)
 
         with patch("voiceio.watchdog.tray") as mock_tray:
             mock_tray.is_alive.return_value = False
@@ -640,7 +640,7 @@ class TestTrayWatchdog:
         vio, _, _ = _make_vio()
         vio.cfg.tray.enabled = True
         vio.recorder.stream_health = MagicMock(return_value=(True, ""))
-        vio.transcriber.healthy = MagicMock(return_value=True)
+        vio.decoder.healthy = MagicMock(return_value=True)
 
         with patch("voiceio.watchdog.tray") as mock_tray:
             mock_tray.is_alive.return_value = True
@@ -651,7 +651,7 @@ class TestTrayWatchdog:
         vio, _, _ = _make_vio()
         vio.cfg.tray.enabled = False
         vio.recorder.stream_health = MagicMock(return_value=(True, ""))
-        vio.transcriber.healthy = MagicMock(return_value=True)
+        vio.decoder.healthy = MagicMock(return_value=True)
 
         with patch("voiceio.watchdog.tray") as mock_tray:
             vio._watchdog.check()
@@ -810,7 +810,7 @@ class TestBootRaceAndReProbe:
         vio._typers.typer = MagicMock()
         vio._typers.typer.name = "clipboard"
         vio.recorder.stream_health = MagicMock(return_value=(True, ""))
-        vio.transcriber.healthy = MagicMock(return_value=True)
+        vio.decoder.healthy = MagicMock(return_value=True)
         vio.cfg.tray.enabled = False
 
         # Env arrives; health check should re-import, re-detect, and upgrade.