Choosing the Right Path When Things Break

OmniGet failures become manageable when the response matches the stage and cause instead of repeating the same command. This chapter turns the system's routing, session, transfer, queue, agent, and refusal boundaries into a practical decision framework. Readers will learn to identify where the workflow stopped, choose the smallest change that could plausibly help, and recognize when waiting, pairing, updating, or changing quality is appropriate—and when the correct action is to stop.

The framework is not a universal or permanent ranking of OmniGet's paths. Native extraction is not always better than generic yt-dlp; a browser bridge is not always necessary; a cloud transcript is not always worth its cost; and more concurrency is rarely a general remedy. Each option is useful under particular constraints.

Diagnose the stage before changing settings

The first question is where the last reliable transition occurred. If OmniGet cannot identify the source or rejects the URL as unsupported, the problem is in routing and stops early. A direct media link, a supported page, or a different source may change the evidence. Changing queue concurrency at this point does nothing because no transfer exists to accelerate.

If preview or metadata retrieval fails, ask whether the platform requires a session, whether the extractor is outdated, or whether the URL identifies a page type the current route does not support. A login wall points toward the authorized browser or managed-cookie path. An unavailable format points toward a less restrictive quality choice or a controlled extractor fallback. A missing local tool points toward setup or update, but installation should remain a user-approved action rather than an automatic side effect.

If the browser extension detected a media item but the bridge is unavailable, the scheme fallback may still enqueue the URL without cookies or manifest text. That can be enough for a genuinely public resource and insufficient for a private one. A missing deep-search manifest is different again: if the page or frame that held the text is gone, retrying a synthetic address cannot restore it. The correct response is new page evidence or a different source, not more queue attempts.

If the item is active, use the queue phase and transfer evidence. A stalled job that is still receiving bytes is not the same as a host returning 429. An HTML body, size mismatch, or selected fatal HTTP status is not the same as a timeout. A merge failure after streams have arrived is not the same as a missing cookie. The same visible symptom—“nothing finished”—can result from several mechanisms.