> During a swipe, the MetroCard is read, re-written to, then check-read to verify correct encoding... Each MetroCard stored value card is assigned a unique, permanent ten-digit serial number when it is manufactured. The value is stored magnetically on the card itself, while the card's transaction history is held centrally in the Automated Fare Collection (AFC) Database... Whenever the card is swiped at a turnstile, the value of the card is read, the new value is written, the customer is let through, and then the central database is updated with the new transaction as soon as possible. Cards are not validated in real time against the database when swiped to pay the fare.
There's also an old NYT article [2]:
> But since the main computer communicates regularly with both the turnstiles and the booth computers, false cards have so far all been invalidated after they have been used once... Some hackers believe that because the turnstiles, the token booth computers and the mainframe computer, respectively, upload and download information at six-minute intervals, the length of time that a passenger can transfer without paying may actually be 18 minutes more than the advertised two hours, if the timing is right... ''I believe that, fundamentally, a system as complicated as the Metrocard cannot be absolutely secure,'' he said. ''What we have to make sure is that the hack won't scale. If a few people ride for free, that's no big deal.''
[1] https://en.wikipedia.org/wiki/MetroCard#Technology
[2] https://www.nytimes.com/1997/08/11/nyregion/what-galls-a-hac...