As per comment#0 this seems to be a feature request. Hence marking this issue as Untriaged.

Could someone from Services>Sync team please have a look at this issue.

Thanks!

Over to mamir who works on USS bookmarks for consideration, though TBH I don't see the connection to USS.

Thx. This is indeed a design/feature request.  
The Title request targeting USS was an uninformed guess.  

Using Google Drive as the bookmark store, in some limited way to scope transaction traffic, would make sense too, since drive provides an archive history.

I'm just looking for a way to deal with the daily user stories of lost bookmarks and no recovery options.

Mohamed:  I scanned the design document
  https://www.chromium.org/developers/design-documents/sync/unified-sync-and-storage-overview

The design addresses performance and IO but there is no mention of recovery, except as 'should not be necessary', nor of recovery requirements coming from the human interface design.

Bookmarks are driven by a human interface.  There will be unintended consequences of correct (as designed) operation, that will require recovery. A redundancy tower of Hanoi model with appropriate period, integrated into Backup and Sync at the bookmark UX level, would be ideal.

From my perspective this is primarily a UX issue but has backend service requirements.

We are currently in the process of migrating bookmark sync code to the latest architecture of Chrome Sync (USS). This will reduce the technical complexity a lot, help to eliminate bugs, and significantly reduce the bookmarks memory footprint. This is our top priority for now.
That effort will not have an effect on the new backup feature you are suggesting. It's worth noting that adding such backup functionality has deeper technical implications (e.g. making sure to avoid merge conflicts) and unfortunately is something we cannot prioritize in the short term.

I can well appreciate the timing and invalidation issues.  

Any minimal backup snapshot, like weekly, would be better than misc. scripts that users have added.  Even if users had to re-import from the backup, that would be easier than the convolutions now required.