CMC on secondaries runs with one of two modes: checkAtPrimaryTimestamp or checkAtSecondaryTimestamp. With checkAtPrimaryTimestamp, the command is sent with local read concern and afterClusterTime to make sure that replication on the secondary has reached the current primary's latest timestamp. The problem with this is that some local reads are performed with snapshot read concern, which creates a potential read skew: local read concern with afterClusterTime only waits for lastApplied to reach the requested timestamp, not the majority committed timestamp, while a snapshot read concern takes the snapshot at the majority committed timestamp, which could be in the past. The solution is to send the command with majority read concern instead, while keeping the afterClusterTime.