If the system or application crashes while a backup cursor is open, (which is true any time a source system does a backup), then when that same source system restarts, it behaves as it would if it were restoring from a backup.  The existing metadata is thrown away and reconstructed from the WiredTiger.backup file, which lists the files in text format.  This seems like an unnecessary step, and contributed to HELP-37618 (WT-9926).

We should consider ways to distinguish the two cases:

• we are starting from a system that was doing a backup, but crashed
• we are starting from a backup directory

Existing applications and organizations (MongoDB, Atlas Cloud, and any direct WT customer), have backup procedures and programs codified, they'll need to still work with whatever we might come up with.

There may not be a satisfactory solution, but it's worth exploring to find out.