Offline Surveys reserves a block of respondent ID's per device per survey. So if device M is the first to get survey A, ID's 1-10XXX are reserved for this device where XXX can be any random number. If another device N downloads survey A then 10XXX+1-20YYY are reserved for it where XXX is the offset for device M and YYY is a random number for device N.
Each device gets a block of slightly more than 10K reserved numbers. These IDs are used when syncing to the server. However, when the device data is merged to the server, if the server encounters a duplicate, it will renumber it server side to an ID that is not currently in use.