The goal with using mcaselector is to avoid needing to either reset the world when it reaches the limit or needing to purchase more storage to allow the world to keep growing.
When I run what we call a 'prune' I use mcaselector to select chunks that are less than X minutes of total Inhabitated time (e.g anything less than 2 minutes for example). This means that chunks that are basically generated while a player is exploring will be marked for deletion as they were not loaded for more than X ammount of minutes, this is usually safe.
Now the problem I faced was that I was finding more and more often people's bases and eventually the spawn was being completely or half reset.. I couldn't understand exactly why. I would keep getting 'deFragment' errors, but eventually found that these are not related.
So the other day my server was down for the entire day due to this and eventually me and a rad staff member came to the conclusion it was the InhabitatedTime that had something wrong... so we checked out some of the problematic selected chunks and found that the time had actually overflowed into a far negatives, so they were being included for deletion because they'll always be below the selected time.
For now we found that we can select all chunks with a InhabitatedTime below 0 (Ticks) and set them to 60 minutes, to avoid them being selected for deletion in the future... We also choose to change any chunk that has gone above the maximum difficulty altering InhabitatedTime back to ... the max difficulty altering InhabitatedTime (3600000 ticks) which will stop chunks from hitting that 32-bit Integer overflow before we run a proper prune with usual settings.
The 'deFragment' errors could be realted to chunks being deleted when they shouldn't be, but the overflow problem could have been drowning out the probelms created by this error, I'll update the post if this is the case.
Not a well written post, but I hope by the end of it this may help someone with the same specific problem.