RAR 5.0 recovery record is based on Reed-Solomon error correction
      codes. If recovery record size is large enough, 5% and more,
      the new error correction scheme provides much higher resistance to
      multiple damages comparing to RAR 4.x recovery record.
      Smaller record, such as 1 - 2%, or less random damage type would
      result in less difference between 4.x and 5.0. For single continuous
      damage 4.x and 5.0 efficiency is about the same.

      Additionally to usual data erasures, the new recovery record
      is able to detect deletions and insertions of much larger size
      than in previous RAR versions. Maximum insertion size is several
      megabytes. Maximum deletion size depends on the damage type
      and in some cases can be as large as the recovery record size.

      Still, the best recovery performance and efficiency is achieved
      if no deletions and insertions are present, so all data including
      damaged sectors preserve their original positions. Thus, if you use
      some special software to copy an archive from damaged media,
      it is better to choose the mode, when damaged sectors are filled by
      zeroes or any other data instead of cutting them out completely
      from resulting file.

      RAR 5.0 recovery record is more resistant to damage of recovery record
      itself and can utilize a partially corrupt recovery record data.
      Note, though, that "Repair" command does not fix broken blocks
      in recovery record. Only file data are corrected. After successful
      archive repair, you may need to create a new recovery record
      for saved files.

      New recovery record is not based on 512 byte sectors anymore
      and incorporates more complicated data structures. So it is impossible
      to specify its size in sectors. For RAR 5.0 archives the parameter of
      -rr[N] switch and rr[N] command is always treated as a percent of
      archive size regardless of presence of % character. Typically N%
      recovery record can repair up to N% of continuously damaged data
      and increases the archive size by only slightly more than N%.
      Ability to fix multiple damages is proportional to N.

      We used "Screaming Fast Galois Field Arithmetic Using Intel
      SIMD Instructions" paper by James S. Plank, Kevin M. Greenan
      and Ethan L. Miller to improve Reed-Solomon coding performance.
      Also we are grateful to Artem Drobanov and Bulat Ziganshin
      for samples and ideas allowed to make Reed-Solomon coding
      more efficient.