Four

New Card Tuesday

Curio Cards had one institution, and it was a day of the week.

“We went towards this idea of printing the cards every Tuesday and every Tuesday we’d have a YouTube show with Travis and I called like new card Tuesday and we would promote the new cards.”

The archive corroborates this more thoroughly than almost anything else in the story, because the Tuesdays left tracks. Hunt told Bitcoin.com in June 2017: “Tuesday is new card day, so we always print new cards on Tuesday.” Both 2017 websites carried a countdown widget labelled “New Card Tuesday!” The front page of mycuriocards.com, captured on 20 July 2017, says simply “Every Tuesday!” above “New Announcements, Limited Supply!” And the tweets are a metronome:

30 May 2017, 17:18 UTC — “RT @MyCurioCards: IN JUST TWO HOURS! We’ve got new #CurioCards! Join us #LIVE at 12:00 PM PDT”

20 June 2017, 16:42 UTC — “Taking the train back to the city to launch New #curiocards! @MyCurioCards”

4 July 2017, 18:08 UTC — “RT @MyCurioCards: Making coffee.... getting ready for NEW #CurioCards Tuesday!”

11 July 2017, 17:05 UTC — “RT @MyCurioCards: Will there be new #CurioCards this Tuesday? Find out in two hours at 12:00 PM PDT”

Three of those four are retweets of the project’s own account rather than Hunt writing in his own voice; the distinction is kept because this book keeps it everywhere else.

Noon Pacific, every Tuesday, a livestream and a drop. It is, in retrospect, a remarkably modern piece of design — the weekly timed release, the stream as the event, the scarcity as the hook — arrived at four years before the NFT market invented it independently and gave it a name.

The last of those tweets, 11 July, contains the seed of the problem. “Will there be new #CurioCards this Tuesday?” By the second month the suspense was not about what the cards would be. It was about whether there would be any.

How it actually worked

The machinery deserves describing properly, because it is the part that survived, and because it is genuinely odd.

There was no NFT standard. This cannot be overstated and Travis Uhrig puts it in nine words:

“They can’t even picture this but there was no marketplace for NFTs. No marketplace.”

Nor much else:

“And we didn’t have MetaMass. There was no MetaMass. There was no Uniswap like Uniswap and MetaMass were the two most biggest forward facing things about Ethereum for a lot of people and neither those things.”

He adds a moment later that MetaMask “technically existed but it was like a beta project”. So: no wallet extension anyone used, no decentralised exchange, no marketplace, no token standard for unique things, no deployment tooling, and no testing framework worth the name.

What existed was ERC-20, the fungible-token standard that every ICO used. Creighton’s solution was to take ERC-20 and abuse it in two specific ways.

One contract per card. Not one contract holding thirty cards — thirty-one separate contracts, each a token in its own right, each named in the pattern Curio1/CRO1, Curio17/CRO17. This is why the card table at the back of this book has thirty-one contract addresses in it. It is also precisely the property that got Curio Cards excluded from the definition of an NFT when ERC-721 was written, and the subject of chapter ten.

Zero decimals. An ERC-20 token normally divides; ether has eighteen decimal places, most ICO tokens had eighteen too. Creighton set decimals = 0. You cannot own half a Curio Card. Hunt describes the reasoning and, in the same breath, the first thing it broke:

“When it comes to locking down the cards to a 1.0 value, which actually broke some of the exchanges like ether delta. They were looking for point ones and point twos and we’re like, no, Curio cards are a card. They are indivisible.”

Uhrig confirms it from the other side, and goes further:

“Gerio found two or three bugs in ether delta because how weird are I don’t know if you remember this there was a bug where ether delta didn’t understand non-divisible cards which is Gerio.”

“Gerio” is Curio. EtherDelta — the only place in 2017 you could trade an arbitrary ERC-20 token, a single-page app wrapped around an on-chain order book — had not anticipated a token that could not be fractioned, and misbehaved when it met one. The misbehaviour was not cosmetic:

“so there was a bug where if you did like a tricky order you could buy a card without buying a card or sell a card without selling a card so you’d get the money and the card or you’d lose the money in the card and people were doing stuff like that”

People were doing stuff like that. And one of the people it happened to was a founder:

“I lost some cards on ether delta because again like our job was adoption right so I took some cards and I put them on ether delta to seed the market you know like how cool and how people can buy them and trade them and they were stolen.”

A cofounder put inventory on the only available exchange in order to demonstrate that a market existed, and the exchange’s handling of his own design decision allowed someone to take them. Those cards are still out there, in whoever’s wallet, indistinguishable from any legitimately bought card. There is no way to know which ones they are.

The vending machine

The second half of Creighton’s design is the part the project named, and the name stuck so well that the word now appears in the official documentation and in the Christie’s catalogue.

A card contract held the tokens. A vending machine was a separate contract that sold them. You sent ether to it; it sent you back the corresponding number of cards. That is all. Hunt’s description of it, live on 20 July 2017, is the clearest contemporaneous account anywhere:

“then you send a theorem from your Curio wallet to the vending machine and the vending machine bounces the card back to your wallet then you can use your public address to check in the gallery to see that you have gotten the card.”¹

Uhrig’s version is more compressed:

“You go you give the contract money and whatever multiple of money to card you gave you that number of cards back.”

And the instructions, from buycuriocards.com — which was, it turns out, a Blogger site, not a bespoke build, with a blog ID and an Atom feed:

“send 0.0025 ETH for 1 / send 0.0050 ETH for 2 / send 0.0250 ETH for 10”

with a reassurance that is pure 2017:

“If the Curio Card you selected is SOLD OUT, your Ethereum will be returned.”

The vending machine is the direct ancestor of what everyone now calls minting — send money to a contract, receive the asset. Uhrig says so explicitly: “that’s now how minting works in the ERC-721 Token Standard.” Whether ERC-721’s authors took it from Curio Cards or arrived at the obvious solution independently is a question this book addresses in chapter ten, and the answer is probably not the flattering one.

There were three machines per card, one per price tier, which is why there were more vending machines than cards. Uhrig, in 2021:

“they found the other vending machines because there were 31 vending machines.”

And the workload that implies, in his own numbers:

“It just was an accidental deploy twice because red had deploy six contracts each week and there were no deployment tools”

Six contracts a week — a card and its machines — deployed by hand, by one person, with no tooling, in weeks where the count sometimes rose to twelve. Hold that number. It explains almost everything that went wrong, including the accident that made the project’s most valuable single card.

MyCurioWallet

The last piece was a wallet, and it existed because of a problem that no longer exists.

In 2017 a custom ERC-20 token did not simply appear in your wallet. You had to add it manually, by pasting in the contract address, the symbol and the number of decimals. For a project with thirty-one separate contracts, that meant asking every buyer to perform the same fiddly operation thirty-one times.

So they forked MyEtherWallet — the browser-based, client-side wallet that everyone used before MetaMask — and shipped a version with the Curio contracts already in it. Hunt, July 2017:

“So if you just scroll down what we have is we have this thing called Curio wallet which is a clone of my ether wallet so it’s a light wallet it doesn’t hold your private keys you just use it to create an account.”

And in January 2018, retrospectively:

“I know when we did a similar project called Curio cards that’s still being worked on, we were using my ether wallet and you were able to add custom tokens to the wallet. the wallet and then later we so one did a fork of my ether wallet that automatically had our custom tokens.”

The official team page credits Kian Bradley as the 2017 Curio wallet developer, which is more than most accounts of the project manage; he is routinely left out.

Uhrig thinks this might have been a first too, and — characteristically — undercuts himself in the same sentence:

“Everyone used my ether wallet and so we we had a fork of my ether wall. I think we’re actually potentially the first people ever to fork my ether wallet as well”

adding that it was “a little bit less cloud on that particular claim”. He means clout. It is the most honest sentence construction in the archive: a priority claim, volunteered and demoted inside twenty words.

The images, meanwhile, were never on Ethereum. They were on IPFS, with the content hash written into the card contract at deployment — visible to this day in the constructor arguments of card 1. Hunt explained it to Bitcoin.com in June 2017 in one sentence that has aged perfectly:

“First we store the image of the Curio Card in IPFS, allowing the images to be hosted on their network. Then we store a link to the image in an ERC20 token on Ethereum.”

That is still, with minor variations, how the overwhelming majority of NFTs work. The project’s documentation now claims Curio’s was the first smart contract to use IPFS to secure artwork. The hash is verifiably in the 2017 call data. The word “first” would require an exhaustive survey nobody has done, and this book does not assert it.

The warehouse, and the burn

Now the strangest design decision, and the one that explains why the supply numbers in this book never quite agree with each other.

Cards 1 to 13 were each minted at one hundred thousand. Not because anyone wanted a hundred thousand, but because nobody knew how many people wanted. Uhrig:

“they thought there were a hundred thousand of some of the cards like cards one through like 13 or something we’ve minted a hundred thousand of them but we never sold a hundred thousand of them that was just like let’s just make a bunch and we’ll figure out how much this all later. And we burned the rest”

Mint a warehouse, sell what sells for a week, destroy the remainder. The project’s documentation describes the same mechanism in cleaner terms:

“For cards 1–10, the supply was decided by the collectors. A large number of Cards were made available, but all the unsold Cards were destroyed at the end of each week. For cards 11–30, the supply was decided by the artist being featured.”

Supply decided by the collectors. It is, genuinely, an elegant idea — demand-discovered scarcity, settled weekly, enforced by burning. It is also the reason card 1 has a supply of 2,154 and card 2 has 1,589 and card 7 has 1,865: those are not design choices, they are sales figures, frozen.

And the project announced the burns as events. There is a second Bitcointalk thread, posted at 21:45 on 16 May 2017, one week after launch, with a title that has no business being as good as it is:

Curio Cards 4-6 now available! Massive burn destroys Curio Cards 1-3 Warehouse!

It opens in the voice of the raccoon — “I’m so glad to have shared my Apples, Acorns and Berries with you!” and “I’m Curio in the City now!” — and offers as proof the balance of the zero address on card 1: 97,846 tokens sent somewhere they can never come back from.

Which means the sell-through on card 1, in its first week, was 2,154 cards. At twenty-five to fifty cents each. Hunt’s tweet the following day:

“Last week we sold more than 5,000 #CurioCards! and now they’re OOP!”

Five thousand cards across three titles. Somewhere between one and two thousand dollars, gross, in the project’s best week ever.

Here is the cost of that elegance. For cards 1 to 13, the on-chain totalSupply is still a flat 100,000, because totalSupply was never reduced — the unsold cards were transferred to an address nobody holds the keys to. So anyone reading the chain naively in 2021 would conclude there were a hundred thousand of card 1. Several people did. Uhrig says the misinformation had to be actively corrected during the rediscovery, and you can see the consequence in this book’s card table, where cards 1 to 13 have a column reading 100000 that means nothing, and cards 14 to 30 have a column where the chain, Christie’s and the project all agree exactly.

Three published totals for the collection exist — 29,185, 29,796 and 29,700 — and none of them reconcile. The burn is why.

A quarter

The price, finally, because it is the number everyone wants and it is in the archive twice.

Hunt, 20 July 2017, waving off a technical question in the same breath:

“so that’s the gas limit for the cards oh I don’t know the technical things the cards cost around 25 cents in American dollars whatever that is.”

And the mechanism behind it:

“for the first three or the first couple sets we were doing different prices we thought that the problem people were buying these is because they weren’t cheap enough right so we would sell like the first hundred for a quarter and the 100 for 50 cents and then the rest of them you could buy for a dollar”

The first hundred at twenty-five cents, the next hundred at fifty, the rest at a dollar. Three price tiers, three vending-machine contracts per card, all hand-deployed. It is a bonding curve, implemented by deploying extra contracts, in 2017.

They stopped doing it, and the reason is the saddest small detail in the chapter:

“eventually we’re like okay you know less work for rent just do the normal dollar pricing we’re just going to standardize”

“rent” is Rhett. They abandoned their cleverest pricing mechanism not because it failed but because it meant one unpaid man had to hand-deploy twelve contracts in a week instead of six.

Robek, on the same stream, supplied the detail that makes the economics look faintly absurd:

“Yeah but the gas limit for the gas limit is like 57k for the for the Vending machine for some reason or 57k is why which is like half of the cost of or like a fourth of the cost of the card.”

Fifty-seven thousand gas to buy a card. A quarter to a half of the card’s price, spent on the privilege of buying it. In 2017 that was survivable. Within two months the network would be so congested that it was not.

They were selling pictures for a quarter in a year when the median Ethereum project raised twenty million dollars. On 26 September 2017, after the Tuesdays had stopped being reliable, Hunt tweeted four words and a handle, to an audience that was not listening:

“Collecting is intensely human. @MyCurioCards”