One of the easiest ways to make a bulk audiobook system look more successful than it is involves a single word: published. A worker reaches the last screen, presses Submit, receives an acceptance response, and the dashboard turns green. The book may still be processing and unavailable to listeners, but the internal job has finished, so everybody starts calling it published.

I ran into this distinction while designing Solti's end-to-end bulk audiobook publishing service. The automation needs to keep moving after a valid submission; waiting for every platform review would waste the worker and stall the catalog. But the business also needs to know what is actually live. Those are two legitimate needs, and one overloaded status cannot answer both.

The practical fix is to keep an action timeline and a publication timeline. Submission belongs to the first. Public availability belongs to the second. They meet at the same platform record, but they are not the same event.

The complication starts after a successful click.

A successful final action proves something useful. It can prove that the platform accepted a submission for an exact record in an exact account. That receipt lets the system release the browser, commit the checkpoint, and work on another eligible title.

It does not necessarily prove that the audiobook is available to buy or stream. Platforms may validate files, process audio, review metadata, surface a needs-action state, or leave the record waiting. Our observed workflows include accepted Google and Spotify submissions whose later public status must be checked separately. That is normal processing evidence, not a reason to erase the submission receipt or declare publication early.

Finished work is not finished processing.

The worker can complete its authorized submission action while the platform still owns the next step. Record both facts without forcing either one to impersonate the other.

This distinction matters more at catalog scale. With one book, a person can remember that “done” really means “submitted Tuesday, still waiting.” With one hundred books, that private translation becomes an unreliable shadow database.

Model two timelines instead of one vague status.

The action timeline answers: what did our system do? It includes preparation, review readiness, the authorized final action, the platform response, the exact record ID, and the committed receipt.

The publication timeline answers: what has the platform most recently shown? It includes Submitted, Processing, Needs action, Live or Published, Rejected, Hidden, Unavailable, and Unknown. The exact words vary by platform, so the system stores both the raw observed value and a normalized internal category.

These timelines should have different clocks. Submission time comes from the recorded submission event. Observation time says when the platform status was checked. If a later check cannot run because authentication expired or the machine is offline, the original receipt remains true while the publication observation becomes stale.

That is better than replacing a known fact with a guess. “Submitted at 2:14 p.m.; last checked at 2:19 p.m.; current observation stale” is operationally useful. “Done” is not.

Require a durable submission receipt.

A browser click is not the receipt. The worker must verify the response against the same record it intended to submit. For Google, that means the same book identity and GGKEY. For Spotify, it means the same UUID and Review or Catalog record. For INaudio, a submission or quality-control state proves submission, not distribution to every retailer.

The durable record should include the internal book ID, profile and destination account, platform, external record ID and URL, submission generation, metadata or artifact revision, timestamp, raw response, and supporting evidence. The worker then reads the committed state back before moving on.

This is the evidence that protects throughput. Once a same-record receipt exists, the worker can process the next book without keeping the browser hostage. It also protects recovery: after a crash, the system can reconcile the original record instead of repeating the final action. That follows the duplicate-safe method in my guide to recovering from uncertain audiobook submissions.

Treat processing as a waiting state, not a failure.

Processing means the platform accepted responsibility for work that has not yet produced publication evidence. The proper response is a scheduled observation, not another upload and not a sleeping worker.

In our current design, planning estimates can help identify unusually long waits, but elapsed time never promotes a title to Published. It also does not automatically prove failure. A beyond-estimate record needs a fresh platform check and perhaps an operator task; it does not need an invented outcome.

The same rule applies when Google shows a post-submission message such as Needs action or not live yet. Preserve the exact message and time. Do not rewrite the submission as though it never happened, and do not pretend the book is live. One record can truthfully contain a successful submission receipt and a current processing or attention state.

Define publication with exact external proof.

Publication is an observation about the external platform, not a reward the internal workflow grants itself. The system must find the exact record in the assigned account and observe the platform state that means Live or Published. Where public availability is part of the requirement, it should also retain the public URL and verification time.

Publication proof has an identity. A matching title in search results is not enough. Similar titles, editions, and account boundaries make that shortcut unsafe. The observed record must match the stored platform ID and expected account, with title and author used as supporting confirmation.

A stale publication observation should remain visible as stale. It should not silently disappear, and it should not certify a new downstream handoff that requires current evidence. “Last verified published yesterday” and “verified published now” are not identical claims.

Gate downstream handoffs on publication, not submission.

The status distinction becomes a safety control at every platform boundary. A Google ebook that is only Submitted cannot authorize creation of its Google audiobook. The audiobook waits for the exact ebook to become Live and for audiobook eligibility to be visible.

A submitted Google audiobook is not yet a verified source for Spotify. The Spotify handoff requires a Live audiobook plus a validated download package, ordered chapter manifest, cover, and artifact hashes. Likewise, Spotify Submitted does not qualify a book for INaudio transfer. The expected source state is Published on the exact Spotify record.

These rules may look conservative, but they prevent downstream work from being built on an assumption. The submission worker continues to the next eligible title; the reconciliation worker watches for publication; the handoff is created once when fresh evidence satisfies the gate.

Make the dashboard tell the uncomfortable truth.

A useful bulk-publishing dashboard should show lane badges, not collapse the entire book into one celebratory label. Google ebook, Google audiobook, Spotify, and INaudio can each be preparing, submitted, processing, published, blocked, or waiting for a prerequisite.

Counts should keep verified submissions separate from verified publications. A bounded run can report that all scheduled submission actions finished while several titles remain in platform processing. That is a successful run with pending publication—not a failure and not a fully published catalog.

I also want last attempted check, last successful verification, account identity, external record link, and freshness visible. If reconciliation is offline, the display should say so. A database refresh may update the screen, but it does not constitute a platform check. I cover that distinction in the platform-monitoring reconciliation guide. The platform remains the authority for publication.

When the status is uncertain, reconcile instead of retrying.

Uncertainty is a separate state because it demands a separate action. If the browser loses connection after Submit, the system does not know whether the click failed or the response was merely missed. Repeating it immediately risks a duplicate.

Recovery begins with the durable intent: exact book, platform, account, record ID, and input revision. The worker searches or opens that same external record and asks whether the intended submission now exists. If it does, the missing receipt can be repaired from platform evidence. If it does not and the platform clearly permits another attempt, the retry uses the same logical action key.

Platform processing similarly schedules a later read. It must never consume an upload retry. Transport errors may receive bounded backoff, but invalid data, wrong-account identity, and unknown final-action outcomes require their own explicit handling.

A practical submitted-versus-published checklist.

For every audiobook lane, I would require these controls:

  1. Name the events separately. Use Submitted, Processing, and Published rather than one generic Done value.
  2. Record the intended identity. Save the internal book, profile, account, platform record, and revision before dispatch.
  3. Verify the final action once. Require a same-record platform response before committing submission.
  4. Preserve raw evidence. Store the platform wording, URL, timestamps, and supporting receipt alongside the normalized status.
  5. Release the worker. A verified submission should not hold a browser lease while the platform processes it.
  6. Schedule reconciliation. Check active records later without repeating uploads or final actions.
  7. Require publication proof. Promote only after the exact record is observed Live or Published in the assigned account.
  8. Gate every handoff. Let fresh publication and artifact evidence—not elapsed time—authorize the next platform.
  9. Show freshness and uncertainty. Keep last success separate from last attempt, and never turn an offline check into current truth.
  10. Report both outcomes. Count completed submissions and verified publications independently for every cohort.

The purpose of this model is not to make the operation look slower. It is to make the operation measurable. Submission proves that our side completed an authorized action. Publication proves that the platform completed its side. Once those receipts are separate, a team can move quickly without lying to itself.

About the author

Mick Southerland is building Solti's managed bulk audiobook publishing operation for owners of finished book catalogs. These field notes document the engineering choices, platform complications, and operating lessons behind publishing at scale.