James,

unspoken assumptions on my side:
routine is planned for field validation during data entry or a batch like run on data entered, but ALWAYS with a human to decide cases where automatic handling fails. That approach gets me out from under the pressure to be 110% correct and helped me a lot in defining rules for identification of duplicates for instance.

Just try to get the automatic percentage above 90% (reduces manual work a lot!) and typically you can get with "easy coding" to percentages >97%. JUst get the code for "human safety net" in order and realize that this is not meant to be highly optimized code - in record validation for this field up to 100ms might be ok and on batch jobs only focus on the "human time" saved<g>.

>>Then look for all occurrencces of substrings consisting only of numbers and strings with 12 numbers.
Here I was talking on a record basis. For all examples you cited "1 12345 12345 1" is the only Subst consisting
of 12 "number" chars and therefore not problematic.

"Partno: 156 597 123 498 Big Widget 1 12345 12345 1"

would be a case where human intervention is needed. Another case would be

"Big Widget 1 12345 12345 1 15 23 54"

but the first step is to find out how many different possible UPC codes are hidden in the string.
As long as it is only one, take that and move on. The fringe cases will include leading zeros and
probably other non-numeric special chars embedded in the UPC code.

UPC_Code("Big Widget 1 00345 12345 1") == UPC_Code("Big Widget 1-345/12345\1") ??

regards

thomas