>Thomas --
>
>Unfortunately, your unspoken assumption doesn't fit my case. This processing actually occurs during a background job, so there is no human intervention possible. Furthermore, there's no date field in the source file that allows me to know which records may have changed, so I must re-process the entire file each time (hourly, as it turns out).

If John's approach works for all your cases and is speedy enough (hourly run...) ok. Depending on other assumptions like
possible insert chars: only spaces or perhaps tabs and other special chars as well
all chars/zeroes put into the meshed fields

perhaps a faster approach could be found "jumping" from the left or from the right. But optimization now would be premature.

regards

thomas