This is a common technique we used when building the data recovery logic. In data recovery, there is a long process to find one correct answer from the read records. Therefore, when processing from the head, it takes time because of the number of steps.
Drives that require data recovery contain a lot of missing information. In other words, if you reconstruct the missing information in advance, you can collect information that can be removed first.
This information is assigned to the retrieved records, removing unnecessary information before processing. Then, instead of starting at the beginning of the search, we randomly select one of the remaining record.
Why select again at random? The amount of computation is negligible, since it is only select randomly. However, it is more likely that the correct record will be included in the remaining records after removing the unnecessary ones, instead of the header you select.
This is best seen with a hundred cards. Mark one of the cards and make it the correct answer. Shuffle and take one. Then remove the 98 unmarked cards. You now have one card and one remaining card. By the way, it is already obvious which is marked. This is the reason for swapping.
