Skip to content

A queued job is not a published post

By ShipContent.

Published November 6, 2026.

Publishing through an agent has three honest results: published, failed, unknown. "Queued" is none of them. On 12 October 2026 we had a Reel live on our own Facebook Page while our job did not record it as published. The send had worked and our verifier did not accept the answer, so a retry at that point would have posted the Reel twice. An agent cannot publish to Facebook in ShipContent. This case is about our verifier and our worker.

Our verifier did not accept a published Reel

On 12 October 2026 we were testing Facebook publishing on our own Page with one text post, one photo and one Reel. Each was approved and sent once. The Reel then went like this:

  1. Meta returned publishing_phase.status=complete and publish_status=published.
  2. Our verifier did not accept that response. It looked for the wrong thing, so the post was live and our job did not record it as published.
  3. The job had a budget of 20 polls and used all 20 before the verifier fix was deployed.
  4. We fixed the verifier and deployed it.
  5. We reopened one read-only verification attempt for the original job. The worker saved Published for the original request, with no second send.
  6. We checked all three posts in native Facebook. The exact captions and media matched.

The failure was on the read side

Meta's answer matched the live post, and the fault was in our verifier. A status from the platform and a record in our database are two separate facts. On 12 October they disagreed, and the platform was right.

Meta's Reels publishing reference documents status with uploading_phase, processing_phase, publishing_phase and publish_status. The last two are the fields in the answer our verifier did not accept.

Only native verification saves the delivery status now

A post counts as published only after the platform confirms it. Our Facebook docs say:

Only native verification saves the final post ID and delivery status. Reel verification requires publishing_phase.status: "complete", publishing_phase.publish_status: "published", and membership in the selected Page's Reel list. A ready upload alone does not confirm publication.

Approval queues one publication job for the worker, and the same page says "a queued or running state is not a provider receipt."

You can publish to a Facebook Page: text, one photo or one Reel. The 12 October results are from accounts the team runs, and no customer has been verified on this path. Your agent cannot do that step in ShipContent. We designed the agent side around one default: your agent prepared and you approved in a signed-in browser. We have no recorded live agent run of that design yet.

Our cross-posting docs give the owner three results:

ResultShown asWhat to do next
PublishedPublished and the timeNothing. The platform confirmed the post.
FailedFailed and the error, with a retry for that platformRead the error, then retry on that platform.
UnknownNot confirmedWait and do not publish again. After 24 hours the owner can select Not live, retry. Check the platform first: if the post is live, a retry posts it twice.

You can copy these four rules into your own agent

  1. Treat queued as not done. Have your agent report a queued job as queued, and report published only when it holds the platform's confirmation. Our Reel shows the same gap from the other side: the post was live and the job had not recorded it.
  2. Do not mint a new request ID to repair a lost response. Our request ID contract says "The ID is a stable UUID for one exact intent, not a retry counter." It also says "A new ID creates a new operation. Never mint a new ID to repair a lost paid generation, provider call, upload, approval, or publication response." A retry after our verifier failed would have posted the Reel twice.
  3. Read the platform before a retry. Our repair was one read-only verification attempt on the original job, with no second send. Give your agent a way to reopen verification alone: our 20 polls ran out before our fix was deployed.
  4. Keep unknown as a real state and show it to the owner. Our cross-posting docs call it Not confirmed and tell the owner: "the result is unknown. Wait. Do not publish again."

We have not proven the failure paths live

The worker is built to mark an uncertain send as unknown and not retry by itself. We tested the failure paths mostly in unit and database tests, not against a live provider failure. After 24 hours an owner can choose Not live, retry, and that can post twice, so check the platform first.

We also built approval to bind the exact content and tested it in integration tests: an edit after the owner opened the review was refused and nothing was queued, and an edit after approval was refused because the draft was locked.

Sources