A file to be written can be partitioned into one or more partitions. Each such input partition is identified with a hash code and a group ID. Replica(s) of an input partition can be created to ensure a certain number of identical partitions. When a file is accessed, each partition that is read out can be checked for corruption. Corrupted readout partitions are then replaced with a validated replacement partition, whether the replacement partition is from a replica or from another file.