Also, when the data file has been modified, and has increased in size, check the consistancy of the last dozen indexes. If they are consistent, only index the new records appended rather than reindexing the whole file.

Actually, checking say log(n) (where n=no of lines) or so randomly selected lines spread across the file for consistancy is probably (probabilistically) sufficient to detect changes, given that any 1-byte change in the length of any line will invalidate the offsets of all subsequent records until an exactly opposite change of length re-syncs them.

Ignoring the possibility of deliberate tampering, any of the maths wizards feel like calculating the odds of a randomly re-written file matching more than say log(n) randomly chosen index offsets from the previous contents consistantly?