Return a validator-safe instruct from a possibly-poisoned stored value. Unlike :func:`_resolve_instruct` (which *raises* on unknown items so the

Top-level entry: list[dict] in, list[dict] out. Preserves every key the caller passed in other than `text`, `start`, `end`, and `words`, whic

Public entry point: whisper result → clean dub segments (as dicts).

Run `_load_model_sync` with a checkpoint load that fails once.

Bring a sidecar up via a health ping and assert it's live.

May the scheduler send work through this breaker right now?

Applies the broadcast pre-stage (highpass + gentle compression) to the clone voice. Reverb is intentionally absent — only user-chosen effect pres

Silence-gap speaker assignment (used when no diarization model runs). Base signal: a gap > SPEAKER_GAP seconds between consecutive segments is

Reserve a unique temporary path for an incoming push. The wire id is HASHED into a directory name rather than used as one. Staged inp

Durably revoke one panel and withdraw all of its live sessions.

Called every step to update the progress bar UI.

Build an ffmpeg filter_complex graph that stretches the source video per-segment so each segment's visual duration matches a dub audio layout.

Fake `openai.OpenAI` with canned chat/models behavior. Returns a list that captures each client's construction kwargs so a test can assert th

Lay out the minimum repo skeleton the validator scans.

Deterministic offline (dub/batch) selection: faster-whisper large-v3, with the HF cache reporting installed/not-installed per ``cached``.

Persist the segments this dub was actually generated from back onto the job. The editor only sends the (translated / user-edited) segment text in

members: {arcname: bytes|str}.

Per-test TestClient; avoids module-scoped fixture collisions with ``patch()`` context managers.

Record an attempt failure and requeue the task if retries remain.

Return true only for a regular, decodable WAV with audio frames.

Measure the concatenated program's loudness for ``preset`` (acx/podcast), or ``None`` for off/unknown or on ANY failure (→ single-pass fallback).

Drop rows for jobs finished more than `older_than_seconds` ago.

Resolve the VoiceStudio TTS checkpoint from ``OMNIVOICE_MODEL``, self-healing a misconfigured value. A valid checkpoint is either a HuggingFa

Run an ffmpeg subprocess with concurrency cap, timeout, and proper cleanup. Returns (returncode, stdout_bytes, stderr_bytes). Raises asyncio.Time

Deterministic RNG seed for one longform synthesis call (#1139). A voice profile's pinned ``seed`` (locked takes, design profiles) makes ``/ge

Finish a durable call and report cancellation after its result is known. Authority writes need their event-loop publication even when the HTTP

Resolve attempt-scoped storage for one result without filesystem I/O. Attempt-scoped, not task-scoped: two attempts of one task must never

Build the ffmpeg command for an audio-only dub export (#119). No video input, stream-map, or codec — just the dubbed track, optionally mixed

Turn the dual gate on. By default pre-marks the install event as sent so tests about OTHER events aren't polluted by the opt-in's install ping.

Fake httpx.Client capturing the probe's URL + headers; GET either returns a canned response or raises.

Internal accessor for the GPU pool singleton. Same object as the module-level `_gpu_pool` attribute, but resolvable from inside this module (P

True when `exc` (or anything in its cause/context chain) is — or carries the text of — the ``RuntimeError`` a ``ThreadPoolExecutor`` raises on

Create a test audio tensor with a simple sine wave.

Stub _probe_available: only the named backend ids report available.

Minimal stand-in for a Starlette Request — the gate only reads client.host.