https://github.com/10gen/mongo/pull/48779/

Test setup. 
Cluster: 3 shards, 1 mongos, 1 config server, balancer off.Setup:

1. Pin DB primary to shard0 (enableSharding with primaryShard: shard0)

1. Create unsplittable collection on shard1 (non-primary) — data lives on shard1

1. Open a collection-level change stream from a cluster time before the commands

Sequence:

1. CreateDatabase → pins primary to shard0 (no events)

1. CreateUnsplittableCollection on shard1 → emits create

1. RenameToNonExistent → emits rename, invalidate

1. InsertDoc (recreates collection implicitly) → emits create, insert, insert

Expected events: [create, rename, invalidate, create, insert, insert]Bug: After the invalidate, v2 resumes using historical placement.

The old namespace's last placement record has shards: [] (renamed away). The pipeline discards it and falls back to DB-level placement → targets shard0 (DB primary).

But the invalidate's resume token points to shard1 (where the data actually was). The resumed cursor opens on the wrong shard and never sees the post-rename create and insert events.

Result: v2 returns only [create, rename, invalidate] — stuck at 3 events instead of 6.