Cloud storage occasionally refuses a request or drops a connection while otherwise perfectly healthy. A single blip like that used to mark an entire dataset as failed, even though the next upload pass would pick it up and complete it moments later.
File uploads now retry automatically, with a short escalating pause between attempts so a busy service is given room to recover. A brief interruption is absorbed inside the upload and never surfaces as a failed dataset. A file is only recorded as failed once its retries are genuinely exhausted, which means the failures you do see are real ones worth looking at.