Operator troubleshooting

Most pipeline failures are transient and recover on their own; the rest you fix from the production detail page with the built-in retry/override actions. This page covers the failure modes you’ll actually see and how to clear them.

🎬 Recover a failed production

Prefer to read? Open the step-by-step transcript

1. Open the production’s detail page and find the step with the ❌ Failed badge.
2. Read the step’s error (code + detail).
3. For Steps 8 (Voiceover), 10 (Video composition), 12 (SEO metadata) or 15 (Gates), click Retry Step for transient causes (AI provider 5xx, timeouts). Every other step needs Manual Override instead — there’s no retry path for it.
4. For exhausted/dead-lettered messages, use the admin DLQ management resubmit after fixing.

Step failures

• Failed status → Steps 8, 10, 12 and 15 have a Retry Step button. Recoverable causes when it applies: AI provider 5xx, timeouts, transient SQL errors. Steps 8, 10 and 12 also offer Refine & Re-run, which lets you change the prompt or model before re-running. For any other failed step, use Manual Override to record how you resolved it and let the pipeline continue — there’s no retry for it.
• Who sees the buttons → the account’s CustomerAdministrator, an assigned AccountOperator and platform admins. A CustomerViewer sees the step’s state but no verbs.
• Dead-lettered → the step exhausted retries. Investigate the error code in the step’s ValidationErrorDetail, then retry (Steps 8/10/12/15 only) or override. (Admins resubmit from Settings → DLQ Management — see Operations & reliability.)
• AwaitingRelease is not a failure — it’s Controlled/Step-through mode holding the step for an operator to click Release Step. See Controlling a running production.

Approved-with-warnings (image safety/rights)

When an image carries warning flags (low safety score, high copyright risk, attribution required), the asset still proceeds, but you must acknowledge the warnings before publishing via the pre-publish rights-review drawer. See Approval and rights review.

Missing assets

If the per-scene image count is below the brief’s planned scene count, the production can stall later in the pipeline (around the platform-variant step). Check the image step’s output on the production detail page and re-run it if needed.

Provider outages

If an AI provider (OpenAI, ElevenLabs, etc.) is in an outage, the affected steps exhaust retries. JARAI routes around a Failing provider using the model-selection chain where it can; check the Status page for current incidents.

Stuck productions

First check whether it’s actually stuck, or just held: a step showing AwaitingRelease is waiting on an operator to click Release Step (Controlled/Step-through mode) — see Controlling a running production. A production that genuinely stalls mid-pipeline (e.g. a message lost to an infrastructure blip) is usually re-driven automatically by the recovery process — give it a few minutes before intervening. If it stays stuck on a step that isn’t one of 8/10/12/15, retry isn’t available; use Manual Override, or Cancel Production if you’d rather start again.

Quick reference

Symptom	Do this
Step AwaitingRelease	Not a failure — release it (see Controlling a running production)
Step 8, 10, 12 or 15 shows Failed	Retry Step (after fixing the cause if permanent), or Refine & Re-run on 8/10/12
Any other step shows Failed	Manual Override — retry isn’t available for it
Step dead-lettered	Admin: inspect + resubmit via DLQ management
Image warnings	Acknowledge in the rights-review drawer before publishing
Many failures, one vendor	Check provider health / Status — the chain should route around it
Production won’t start	Check the account’s budget/allowance and concurrency limits
Production stuck (not AwaitingRelease)	Wait for auto-recovery; if persistent and on Step 8/10/12/15, retry — otherwise, Manual Override or contact support
Production waiting in PendingApproval	Not stuck — click Approve distribution on its page (auto-publish is off for the account)

Caution

Don’t manually edit the database or storage to “fix” a failed production — JARAI’s reconciliation will detect the drift and may revert your changes. Always use the Console’s retry / override actions.

Still stuck? See the FAQ or Contact support.