The photo that verifies itself
A customer texts a licence from their couch. Ninety seconds later you have a scored identity verdict and a PDF for the file. Here is what has to be true for that to be trustworthy.
Identity verification in most stores happens at the desk, on the day, by a person looking at a card. It works, mostly. It is also the latest possible moment to find out something is wrong, and it only happens with real scrutiny when somebody already feels suspicious.
Doing it from a texted photo, on every deal, changes both of those. But only if the check is actually rigorous, and most automated document checks are not.
What a real check has to do
Read the barcode, not the picture
The barcode on the back of a licence is the authoritative record. It is what a police scanner reads. Everything printed on the front is a rendering of that data, and it is the part that is easy to fake.
A check that reads only the front is reading the least trustworthy surface of the document.
Cross-check, do not just extract
Having the name from the front and the name from the barcode is not the point. Comparing them is the point. Same for date of birth, expiry, and — for Ontario licences — the date of birth encoded inside the licence number itself, which is a check most people do not know exists.
Score, and show your work
A verdict of verified with no reasoning is not usable in a compliance file. What you want is a score with explicit deductions: no readable barcode, expired, licence number mismatch, implausible age, unreadable front. Then a human can see why, and disagree if they want to.
The rule that makes the whole thing trustworthy
If it cannot read the barcode, it can never return a pass. The best available outcome is a caution telling a human to look at it.
This sounds like a limitation. It is the opposite. Any system that will grade a blurry front-only photo as verified is a system that a determined person can use to launder bad ID into your deal jacket, and its passes are worthless because you can never tell which ones were real.
A ceiling on what an incomplete capture can achieve is the thing that makes a complete capture mean something.
The operational payoff
Two things change. You find bad ID before it is in the file rather than at the desk. And you run the check on every single deal instead of only the ones where somebody felt uneasy, which is where the real exposure always was — nobody feels uneasy about the well-dressed customer with the confident story.