One of the chronic features of the Bitcoin landscape is that Bitcoin exchanges screw up and fail, starting with Mt. Gox. There’s nothing conceptually very hard about running an exchange, so what’s the problem?

The first problem is that Bitcoin and other blockchains are by design completely unforgiving. If there is a bug in your software which lets people steal coins, too bad, nothing to be done.

Some environments need software that has to be perfect, or as close as we can make it, such as space probes that have to run for years or decades, and implanted medical devices where a bug could kill the patient. Programmers have software design techniques for those environments, but they generally start with a clear model of what the environment is and what sort of threats the device will have to face. Then they write and test the code as completely as they can, and burn it into a read-only memory in the device, which prevents deliberate or accidental later changes to the code.

Running an online cryptocurrency exchange is about as far from that model as one can imagine. The exchange’s web site faces the Internet where one can expect non-stop hostile attacks using rapidly evolving techniques. The software that runs the web site and the databases is ordinary server stuff, reasonably good quality, but way too big and way too dynamic to allow the sorts of techniques that space probes use. Nonetheless, there are plenty of ways to try and make an exchange secure.

A bitcoin exchange receives bitcoins and money from its customers, who then trade one for the other, and later ask for the results of the trade back. The bitcoins and money that the customers have sent stay in inventory until they’re returned to the customers. If the exchange closes its books once a day, at that point the bitcoins in inventory (which are public since the bitcoin ledger is public) should match the amount the customers have sent minus the amount returned. Similarly, the amount in the exchange’s bank account should match the net cash sent. The thing in the middle is a black hole since with most bitcoin exchanges you have no idea where your bitcoins or cash have gone until you get them back, or sometimes you don’t.

To make it hard to steal the bitcoins, an exchange might keep the inventory in a cold wallet, one where the private key needed to sign transactions is not on any computer connected to the Internet. Once a day they might burn a list of bitcoin withdrawals onto a CD, take the CD into a vault where there’s a computer with the private wallet key, create and sign the withdrawal transactions, and burn them onto another CD, leave the computer, the first CD, and a copy of the second CD in the vault, and take the second CD to an online computer that can send out the transactions. They could do something similar for cash withdrawals, with a bank that required a cryptographic signature with a key stored on an offline computer for withdrawal instructions.

None of this is exotic, and while it wouldn’t make anything fraud-proof, it’d at least be possible to audit what’s happening and have a daily check of whether the money and bitcoins are where they are supposed to be. But when I read about the endless stories of crooks breaking into exchanges and stealing cryptocurrencies from hot (online) wallets, it’s painfully clear that the exchanges, at least the ones that got hacked, don’t do even this sort of simple stuff.

Admittedly, this would slow things down. If there’s one CD burned per day, you can only withdraw your money or bitcoins once per day. Personally, I think that’s entirely reasonable—my stockbroker takes two days to transfer cash and longer than that to transfer securities, but we all seem to manage.