Pagination, retries, and checkpoints

APIs rarely hand you the whole dataset in one response. They hand you pages, cursors, rate limits, and the occasional retry.

A useful ingestion job remembers the last page it safely wrote. On retry, it resumes after that checkpoint instead of starting from scratch or duplicating a page. The checkpoint is part of the dataset evidence, not a private implementation detail.

The worker move is "the script finished on my machine." The builder move is "the next run knows exactly where to resume."