Is there any mechanism to check the data that is replicated in recovery site is consistent or not? - Yes, see below.
It does change block tracking evey time (Subsequent Replications) by using block by block comparison of source and target disk or it just refers .psf file? - it doesn't use CBT. It uses a LWD (which is similar to but not the same as CBT). It doesn't do a block by block comparison on an ongoing basis. The vSCSI filter tracks writes as the VM writes to it's disk.
See below for details:
A lightweight snapshot is a consistency group across all virtual disks in a VM. When a snapshot is created, all writes are “synchronized” across all virtual disks using a Two-Phase-Commit protocol in the vSCSI filter. This provides write order consistency across all virtual disks. The replica data is not written to the shadow VMDK until entire block group has been received
vSphere Replication creates a redo-log on the disk at the target location. When a replication takes place the vSphere Replication Appliance receives the changed blocks from the source host and immediately writes them to the redo log on the target disk. When the complete replication bundle (the “light weight delta”) had been received, it then commits the redo log to the target VMDK file.
VR then retains the redo log in the form of a snapshot, which is automatically committed during failover. This snapshot is always retained in case of error during the commit; during crash or corruption it ensures there is always a ‘last known good’ snapshot ready to be committed.
