SERVER-98386 fixed a bug caused by a timeseriesView->buckets translation done when setting the ShardRole OperationShardingState in the service entry point.
That same pattern also exists on bulk_write.cpp and it was not addressed by SERVER-98386 – the ShardVersion requested by the router for a given nss, say {nss1 -> SV1} is translated to {db.system.buckets.nss\ -> SV1}. This makes the ShardRole collection acquisitions for nss disregard the ShardVersion requested by the router, which may mean we fail to detect stale targeting. The bulk_write case appears to be benign – as in, there's no specific correctness bug caused by it, because bulk_write.cpp is only used for writes, and the bug reported on SERVER-98386 exclusively affected reads. Regardless, we should fix the bulk_write.cpp code to make it safer and easier to reason about.