It's now 2026, and I had AI go back over the numbers I mentioned at around the 2:30 mark. Here's the clarification.

For two sales reps adding a few dozen or a few hundred records while they're away, random AutoNumbers still make a collision very unlikely. The practical example is reasonable. However, my one-record-every-second-for-a-year comparison needs some context.

Using a simplified model of about 4.3 billion equally likely numbers, here is the chance that two separate copies generate at least one matching ID:

100 new records in EACH copy: about 1 in 430,000.
1,000 new records in EACH copy: about 1 in 4,300.
10,000 new records in EACH copy: about 1 in 43, or 2.3%.

These are calculated odds for that model, not measurements of Access's random-number generator. They count new records created independently after the copies separate, not records they already share.

The "1 in 136" figure describes ONE additional random number matching one of roughly 31.5 million existing numbers. It does not describe the accumulated chance of a conflict while two separate databases keep adding records.

Bottom line: very unlikely is not the same as impossible, and the risk grows as you add more records or more independent copies. For a small team entering modest numbers of orders, this doesn't mean you need to throw out your database. It does mean a merge should flag unexpected ID conflicts for review rather than assume every matching ID is the same record.