Long documents take a while to draft. A dropped connection or a closed laptop lid shouldn't mean starting over.
Every xDocs generation job is a sequence of named checkpoints — plan chapters, draft each one, assemble the manuscript, write the version. After each stage completes, the job records exactly where it left off. If a worker crashes or a network drops mid-chapter, the next attempt resumes from the last completed checkpoint, not chapter one.
Because each stage is idempotent, a retry never re-runs (and never re-bills) work that already succeeded. You only ever pay once for a given chapter, even if the underlying job needed a second attempt to finish.