Seven

Two Seventeens

The most valuable single thing Curio Cards ever produced was a mistake, and nobody involved can tell you with confidence what the mistake was.

Here is what is not in dispute, because it is on the chain. Two contracts exist, both named Curio17, both with the symbol CRO17, both with a totalSupply of 500. One sits at 0xF59536290906F204C3c7918D40C1Cc5f99643d0B and was created on 20 June 2017 at 21:50:49 UTC. The other sits at 0xE0B5E6F32d657e0e18d4B3E801EBC76a5959e123 and was created on 20 June 2017 at 07:05:52 UTC, in block 3902088, through a different factory contract at 0xeca65Be784e2B0F1956D266c1237481a511a19FB.

Read those timestamps again, because every retelling gets this backwards. The card the world calls the misprint, card 17b, went up roughly fourteen hours and forty-five minutes before the card it is supposedly a misprint of. The duplicate came first.

Two consequences follow immediately. The first is that the misprint is a June artefact, not a May one — it postdates the first CryptoPunks contract by eleven days, which matters to anyone trying to use 17b in a priority argument. The second is smaller and stranger. At 16:42 UTC that day, between the two deployments, Hunt tweeted:

“Taking the train back to the city to launch New #curiocards! @MyCurioCards”

By the time he sent that, the thing that would become the collection’s rarest tradeable card had already been sitting on Ethereum for nine and a half hours.

One correction to the record while the addresses are in front of us. Christie’s lot listing gives the same contract address for both card 17 and card 17b — 0xF595… for each. That address is card 17’s. The auction house that sold the set including the misprint published the misprint’s identifier incorrectly.

What the accident was

Four accounts exist. They do not agree, and this chapter does not pretend otherwise.

Hunt’s version is secondhand and he says so. On Adventures in #NFTs:

“I think ret told me something like I was late night I screwed up the contract I had to do it twice”

Note the construction — I think he told me something like. Two hedges and a paraphrase in fourteen words. Hunt is reporting a years-old recollection of a thing someone else said about his own project, and he marks it as exactly that.

He said the same thing in a different register six days after the rediscovery, on The Bitcoin Group #250 on 19 March 2021:

“there’s actually a fork on card 17. 17 the UASF card. I don’t know what happened. My theory is that ret made mistake and then fixed it but there are two different versions”

I don’t know what happened. My theory is. And again on 7 June 2021:

“as i recall uh rat made an error uh there was something wrong with the contract and he had to do it twice so there became two of these cards”¹

Travis Uhrig’s version is that he did not know 17b existed. This is the sharpest contradiction in the whole history, because he was the founder who kept the project alive through the quiet years:

“Card that I didn’t know about they’re like hey what about the 17 B and I’m like there’s no 17 B what is that you’re getting scam.”

He went further than disbelief — he told people not to buy it:

“But no and I actually thought for why I told people like don’t buy it because there must be something wrong with it right the only reason why we would have deployed it twice is if there was a mistake.”

That reasoning is correct and the conclusion was wrong, which is the most human thing in the archive. The only reason to deploy twice is a mistake; therefore the duplicate is defective; therefore do not buy it. The market disagreed and 17b became an essential.

His first guess about the nature of the error was something else entirely, and it opens a door nobody has walked through:

“My first guess was we used the wrong IPFS hash because I remember that did happen there was a card that was deployed at the wrong IPFS hash but that would never went up for sale so it doesn’t matter.”

A second, separate 2017 incident — a card deployed pointing at the wrong image, which never went on sale — is documented nowhere but that sentence. It is not 17b and it is not in any published account. The apparatus lists it among the things that could not be established.

Then Uhrig supplies the detail that actually explains the whole affair, and it is under-reported everywhere:

“And 17 wasn’t the only one that was deployed twice because of the mistake or or whatever a couple of them were deployed twice but 17 was the only one that the machine is also deployed those only when you actually buy the other ones are just inaccessible.”

Several cards were double-deployed that week. Only 17’s duplicate also got a working vending machine. That is the technical reason 17b exists as a card you can own and the other duplicates do not — not that 17 was special, but that its spare had a cash register attached.

He is not consistent on this across years, and the book shows the drift rather than picking the best take. In 2023 he says 17 was not the only one deployed twice. In 2024 he says “it was 18 and 19 … 17 was the one that was not messed up”. Elsewhere in 2023: “Actually 1718 and 19 were all double-manated, but only 17 was valuable”. The three versions converge on all three cards being double-deployed with only 17’s duplicate functional, and the 2024 version explicitly throws out the poetic reading this chapter is about to arrive at. That is noted here before the poetry, not after.

Uhrig’s published version is the clearest of all. Speaking to CoinDesk on 17 September 2021, two weeks before Christie’s:

“Cards 17, 18 and 19 were all initially misprinted, but there was a mistake that made 18 and 19 unsellable,” explained Uhrig. “Back in 2017, nobody cared that you had test code on Ethereum. We just re-deployed all three after fixing [the error]. People eventually found [card #17b] because it lives on the blockchain. The market has definitely spoken, people think of it as an essential in a full set.”

The square brackets are CoinDesk’s, not this book’s.

The project’s own documentation gives the fullest account, and it is consistent with Uhrig’s published one:

“Lack of testing frameworks and low gas costs made “testing on mainnet” common practice at the time of the project’s development. When Cards 17, 18, and 19 were deployed and tested, an issue with the vending machines was found that made Cards 18 and 19 unpurchaseable. The issue was fixed, all of the week’s contracts were redeployed, and Cards 17–19 were officially released.

However, the first 17 Card still worked. Its vending machine wasn’t affected and still allowed purchasing. At the time, it wasn’t listed for purchase, and was simply forgotten about; a leftover artifact of a standard test deployment.”

So: the morning deployment was a test. Testing on mainnet was normal because gas was cheap and testing frameworks did not exist — the point chapter four makes about what 2017 lacked. Cards 18 and 19 came out broken, everything was redeployed that evening, and the morning’s card 17 was simply left there, working, unlisted, forgotten. It explains the fourteen-hour-forty-five-minute gap exactly: a test run at seven in the morning, a fix, a real run at ten at night.

Adam McBride’s version contradicts the documentation, and the chain decides against him. Speaking to Amy Castor and David Gerard in June 2022:

“I don’t think anyone knows exactly what happened with 17b, but it was briefly released,” McBride said. “One person who is actually in the Discord was able to get, I don’t know how many 17b’s, before they realized their mistake and turned off the contract.”

His first clause is fair and this chapter agrees with it. The rest does not survive contact with the ledger. If the 17b contract had been switched off in 2017 after one person bought an unknown quantity, 17b could not have ended up with 497 of its 500 cards sold. The vending machine stayed live for four years, which is the entire reason there was anything left to find in 2021.

Did Hunt know?

This is the contradiction the book cannot resolve and says so.

Hunt’s “I think ret told me something like I was late night I screwed up the contract I had to do it twice” reads as a man recalling being told about the problem at the time, in 2017. Uhrig’s “there’s no 17 B what is that you’re getting scam” is a man learning of the duplicate’s existence in 2021, from strangers, and refusing to believe it.

Both can be true — Creighton could have told one founder and not the other, on a late night, about a test deployment that seemed too trivial to circulate. Neither account is corroborated. Creighton has made exactly one public statement about Curio Cards in nine years and it is not about this. The documentation’s “simply forgotten about” is compatible with both and settles neither.

Unresolved. It is listed as C3 in the back of the book and it will probably stay there.

The card about the fork, which forked

Card 17 is called UASF. The artist is Cryptopop — Luis Buenaventura, who had flown in from the Philippines and been introduced at the San Francisco meetup, and who would later also make the artwork used for the 17b wrapper.

UASF stands for User Activated Soft Fork. In the summer of 2017 it was the name of a campaign in the Bitcoin scaling dispute, and June 2017 — the exact month card 17 was deployed, twice — was the peak of that fight. A card commemorating a proposal to fork a blockchain was itself accidentally forked into two contracts on the day it was made.

Everyone involved has noticed. Hunt:

“but during the readiscence of these key or cards this 17 card was discovered as this alternate contract and it couldn’t have happened on a better card right because this is u a s f”

And:

“and just like in our cards buy I think lucky accident there were two 17 cards”

Uhrig, in 2021, agreed:

“so it’s really just kind of a happy accident that reds error came up on this the fork card.”

“reds” is Rhett’s. It is a good coincidence and this book enjoys it as much as anyone. But Uhrig himself walked the reading back in 2024, pointing out that cards 18 and 19 were double-deployed too, and that the fork card’s fork was therefore one of three accidents that week rather than a singular stroke of fate. He is right. The coincidence is that the surviving duplicate landed on the fork card — and the reason it survived is a vending machine, not destiny.

What it is worth being wrong about

The commercial epilogue is the part that should bother anyone who thinks markets price information.

Card 17’s on-chain supply is 500; the project’s own site reports 492 sold. Card 17b’s on-chain supply is also 500, with 497 sold. The two cards are, as objects, identical: same name, same symbol, same supply, same image, same artist, one of them deployed by accident in the morning.

A full set of Curio Cards is now understood to require thirty-one cards rather than thirty, because the accident is considered essential. Christie’s sold the set as “Full set of 30 cards plus 17b”. Uhrig’s own phrase for it — “The market has definitely spoken” — is the correct one, and it is worth noticing what the market spoke about. It assigned value to a deployment error on the basis that it was an authentic deployment error, verifiable on chain, which is a thing you can only do if the error was never cleaned up.

Nobody cleaned it up because nobody could. The contract had no owner function to disable it and no version number to supersede it, and the reason for that is the single most consequential technical fact about Curio Cards, which none of its founders can explain. Hunt:

“But for whatever reason when he sat down to program it, he didn’t include version numbers”

The property that made the cards permanent is the same property that made the mistake permanent. In 2017 that was a liability — a broken test card loose on the main network, with a working till, that the team could not recall. By 2021 it was the most sought-after single item in the collection.

The misprint is not an aberration in this story. It is the story, run in miniature: a thing done carelessly and cheaply in a week when nobody was watching, preserved by accident because the system had no mechanism for second thoughts, and revalued four years later by people who had no idea how little care had gone into it.