next paging, the status fields, errors, maxTextLength and webhooks work as in Read a crawl, at /v1/batch/scrape. A batch and its pages are kept for 24 hours, then reading it returns 404 notFound. These calls are not charged.
404 or 410 counts as completed, with metadata.error and no charge. failed counts pages that could not be read. Each page in data is a scrape answer.
Calls
The older
creditsUsed field is a legacy field. Use costUsd.
Retry failed pages
Read/errors and take the url of each row with retryable: true. Send them in urls to a new POST /v1/batch/scrape with the same options, or to the running batch with appendToId. Wait a few seconds before retrying sourceUnavailable and sourceRefused. Do not retry a row with retryable: false, such as notFound, siteNotRead or promptInjectionDetected. Codes are in Errors and retry timing in Troubleshooting. Batches share job limits with crawls: see Rate limits.