Clarification required on the AWS Rekognition bulk analysis

Amazon Rekognition's Bulk Analysis feature processes large collections of images asynchronously, but the documentation doesn't explicitly state whether the output is written incrementally or all at once.

What we do know is that the bulk analysis job generates an output manifest file containing the job results, along with a manifest summary that includes statistics and details about any errors encountered during processing. The results of a finished job are outputted to a specified Amazon S3 bucket that you configure when starting the job.

When you initiate a bulk analysis job using the StartMediaAnalysisJob operation, you specify the S3 bucket and optional prefix where you want the results to be stored. You can then retrieve information about the job, including the S3 path where results and summary files are stored, by using the GetMediaAnalysisJob operation.

The output manifest contains entries for each processed image, formatted as JSON lines. For example, with content moderation, each line would include the source reference to the image and the moderation labels detected (if any).

Based on the available documentation, it appears that the complete results are made available once the job finishes processing all images, rather than being written incrementally as each image is processed. However, the documentation doesn't explicitly confirm this behavior.
Sources
Bulk analysis - Amazon Rekognition
Bulk analysis output manifests - Amazon Rekognition
Processing images in bulk - Amazon Rekognition