Before retrying an interrupted generation task, inspect its actual status and any saved result. Match recovered output to the request, chapter, and source version, keep it separate from approved text until reviewed, and retry only after the existing attempt’s state is understood.
Separate an observation failure from a generation failure
A writing workflow can lose its view of a request without proving that the request itself failed. An interface may stop updating or a connection may end while another part of the process continues. Begin recovery by checking the actual status information and saved artifacts available for that specific attempt.
Do not launch a replacement solely because the waiting screen disappeared. Likewise, do not assume a result exists because the task once showed progress. Record what is known: a live request, a completed result, an explicit failure, or an unresolved state. The next action should follow that evidence rather than a guess based on elapsed time.
Match the saved output with its request
If a draft file or response is found, identify the request that produced it. Compare the chapter assignment, input manuscript version, generation settings, and any recorded task identifier. A file named chapter draft is not enough to establish that it belongs to the interrupted attempt currently being recovered.
Keep unmatched results separate while investigating. A plausible passage may concern an older outline or a different chapter with a similar title. Importing it immediately can create a subtle continuity problem that looks like ordinary editorial weakness later. Establish the relationship before deciding whether the output is useful for the current manuscript.
Check whether the result is complete
Inspect the recovered content for a coherent ending, required sections, and any indication that the response stopped partway through. Compare it with the assignment rather than accepting a file's existence as completion. A saved fragment can be valuable evidence without being a finished chapter.
Record the result's status precisely. It may be complete but unreviewed, partial and recoverable, or unsuitable for the current source version. Keep those states distinct from approved. A recovery process should preserve useful work and make its limits clear, not promote uncertain content merely to clear the task queue.
Protect approved edits made during the interruption
The manuscript may have changed while the request's state was unresolved. Compare the recovered draft with the current approved chapter before merging anything. A result based on older inputs should not overwrite an editor's later corrections, even if the request was originally authorized to replace that chapter.
Use a review copy or a controlled comparison to select useful passages. Record which version was the basis for the recovered result and which changes were accepted into the current manuscript. If the differences are too extensive to reconcile safely, retain the old result as history and plan a new task from the current approved inputs.
Decide whether a retry is still necessary
If the attempt completed and the result meets the task's structural requirements, move it to the normal editorial review instead of generating the same chapter again automatically. If it failed or the recovered content is unusable, identify the reason before retrying. A missing input, outdated brief, or unavailable provider balance will not be repaired by repeating an unchanged request.
Keep the retry linked to the earlier attempt while giving it a distinct identity. This makes duplicate results recognizable if an older response later appears. The workflow should allow the author to understand which result was selected and why, rather than leaving several similar drafts competing to become the current chapter.
Close recovery with a traceable manuscript state
After reconciliation, record the selected output, any rejected or partial result, and the current chapter revision. Confirm that the queue reflects the actual remaining work. A task can be complete as a generation operation while still needing factual, stylistic, and publication review.
Preserve the origin of any generated text included in the manuscript and apply current KDP content guidance during disclosure review. Recovering a response or editing it extensively does not erase how it was created. The safer continuation point is a clearly identified draft with known inputs and review status, not simply a workflow that has stopped showing an error.
Frequently asked questions
Clear answers for this publishing decision.
Does a timeout mean I should regenerate the chapter immediately?
No. Check the specific attempt’s actual status and any saved output first. An observation timeout does not by itself establish that generation failed.
Can a recovered draft replace the current chapter automatically?
Compare its source version with the current manuscript and protect later approved edits. Keep the result unapproved until its identity, completeness, and suitability have been reviewed.
What if both the old attempt and the retry return usable drafts?
Keep both identities and review their relationship to the current assignment. Select or reconcile deliberately, record the decision, and prevent either result from silently overwriting the accepted version.
How wide does the full-wrap cover need to be, and how thick is the spine?
The cover width is back cover plus spine plus front cover plus bleed on both sides; the spine width is the final even page count multiplied by the paper-specific factor Amazon publishes, so the cover cannot be sized before the interior page count is final.
Prepare the cover in a book project