I was trying to reduce the problem down to only the parts that matter. Through out the message, there are 10 markers - I am extracting and only interested in 9 of them. It is weird and highly domain specific (which I can't discuss).
This "set" of numbers for a given message is valid over a few days (let's say 10 days tops). The reset happens when the sender's right to send expires and they are removed so it is possible for the same numbers to be re-used at some point in the future but they would be considered "new" then. At any one time, I think I would have around 500K of the nearly 900 million possible.