As described by milkie:

In this failure, an index on test13_fsmdb0.create_capped_collection_maxdocs0_2 is validated successfully multiple times, but at some point, validation fails (only on the primary node), with three dangling index table records with no corresponding collection table records. Presumably these index records were deleted (in transactions that also deleted their collection table counterparts), but came back into visibility at some point afterward.

This is related to capped collections, so we should check whether the data visibility issue is related to maintaining indexes on secondary indexes and using mixed mode timestamp writes.