Bitcoin got two and a half times busier. BIP-110 would not have caught what did it.

In December 2023 a person sending bitcoin paid roughly nine times what the same payment would have cost a year earlier. The coin had not become harder to move. The transaction was the same size, went to the same sort of address, and did the same job. What had changed was who else wanted the room.

The competition had a name. In early 2023 people worked out how to store arbitrary files inside Bitcoin transactions, pictures mostly, using a scheme called ordinals, and the files themselves became known as inscriptions. Within days they were taking up a third of every block.

Bitquery indexes every Bitcoin transaction ever made and makes them queryable through an MCP server. We took four years of them, from before inscriptions existed through to this month, and measured what they did: how much room they took, what they cost the people paying to send money, what they left behind in the permanent memory of every computer running the network, and what miners earned from them. Then we checked whether any of it is still happening, because this month miners were asked to approve a ban on it. Support collapsed. What the measurements show is that it was aimed at something which had largely stopped, and would have missed almost all of what took its place.

What a block is, and why people fight over it

Bitcoin gathers up transactions and seals them into a block about every ten minutes. Each block holds a fixed amount, and that amount cannot be raised without near-unanimous agreement from everyone running the software. When more people want to send money than there is room for, they bid against each other. Whoever offers more per unit of space goes in sooner. Everyone else waits.

The bidding is priced by size, not by value. Moving a fortune costs the same as moving lunch money, provided the two transactions take up the same space. So the price is quoted the way freight is quoted, per unit of room: satoshis per virtual byte. A satoshi is the smallest slice of a bitcoin, one hundred-millionth.

There is also a detail from Bitcoin's past worth knowing. In 2017 the software was changed so that signatures moved into a separate compartment of each transaction, called the witness, and that compartment was given a discount. Space inside it counts a quarter as much toward the block limit as space anywhere else. The point was to make ordinary payments cheaper. The side effect was that anybody who found a way to put something other than signatures in there would be buying blockspace at a quarter price.

In early 2023, somebody did.

The first week of February

The witness had been filling up slowly for years as wallets adopted the newer format, from about a twentieth of each block in 2018 to about a sixth by the end of 2022. Five years, twelve points. Then it moved differently, and fast enough to date precisely.

The chain got heavier at the same time. Every computer holding a full copy of Bitcoin stores every block forever, so this is a bill that goes out to volunteers, and it rose by more than half in a single month.

What it cost to send money

Fees did not jump straight away. Through February the price of sending money sat about where it had been all winter. Blocks were fuller, but not yet so full that ordinary payments had to outbid anything.

That came later, and it came hard. To see it properly you have to separate people from machines, so we isolated transactions that look like somebody moving money: a handful of inputs and outputs, ordinary signatures, and no data attached.

Two things sit in that table. The obvious one is the middle row, where the typical payment cost fourteen times what it had before inscriptions, and the worst-hit tenth were paying rates that made small transfers pointless.

The quieter one is the first column. The number of payments barely moves across four years. Before inscriptions, Bitcoin carried about seven million of them a month. It carries about seven million now. Everything else on the chain arrived on top of that, which is why the share column falls through the floor while the count stands still.

There is a fair objection to make here, and it is worth making. A flat count does not prove that Bitcoin is used no more than it was. Payments may have moved somewhere this measurement cannot see: onto the Lightning Network, which settles to the chain only occasionally, or inside exchanges and funds, where a customer buying bitcoin produces an entry in a company database and nothing on the chain at all. Both have grown. Neither leaves a trace we can count here.

What the chain does say is narrower, and still worth knowing. Of the transactions Bitcoin itself carries and charges for, the share that moves money between people has fallen from roughly nine in ten to roughly one in three, and the absolute number has not grown at all. We have counted a chain this way once before, on Polygon, and found much the same shape: a network whose activity figures describe one application rather than a population.

The part that never goes away

Blockspace is rented. You pay for it once, the block is sealed, and the network moves on.

There is a second cost that does not work that way. Every parcel of bitcoin sitting at an address waiting to be spent has to be held in the working memory of every computer that checks the network's rules. Collectively these parcels are called the UTXO set, and it is one of the things that decide how much machine a person needs in order to participate. The set only shrinks when coins are spent, and plenty of what goes into it will never be spent by anybody.

Before inscriptions the set was flat, and in some months it shrank. In December 2023 it grew by more than fifteen million entries. That is the durable part of the bill, and unlike the fees, nobody gets to stop paying it.

One smaller practice deserves naming here because of how much permanent state it buys with very few bytes. Stamps encode data by disguising it as a locking condition on a coin, which produces outputs of roughly eight hundred satoshis that are, for practical purposes, permanently stuck. They peaked at just under two hundred thousand of these a month in January 2024.

What miners got out of it

Miners are paid twice. There is a subsidy of newly created coins, fixed by the software and cut in half every four years, and there are the fees people pay to get into blocks. The subsidy is on a schedule that ends. Whether fees can eventually replace it is the long argument about how Bitcoin pays for its own security, and the share of miner income that comes from fees is how people keep score.

For one month at the end of 2023, fees were more than a fifth of what mining paid. That is the closest Bitcoin has come in years to the future its designers sketched, and it was bought with the most expensive blockspace ordinary users had seen since 2018. It lasted a season. Today the fee share is back under one percent, and the chain is busier than it was then.

It is worth saying that the 2023 squeeze was not the worst on record. In January 2018 the typical payment cost about three times what it cost in December 2023. What made the later episode different was the cause. In 2018 blockspace was scarce because a great many people wanted to move money at once. In 2023 it was scarce because of something that was not money at all.

It stopped, and something else started

Inscriptions are largely finished. Transactions carrying a big lump of witness data now account for a fraction of a percent of what Bitcoin does, and a couple of percent of the space.

The blocks are still full. What fills them is a token claim.

Roughly six in ten Bitcoin transactions now carry a small marker belonging to Runes, a token system that sits on top of Bitcoin. The overwhelming majority of those markers are claims on one token, UNCOMMON•GOODS, which was written into the protocol itself at the 2024 halving with no presale and nothing set aside for anybody. Anybody can claim a single unit per transaction, and the window to do it runs until 2028.

So people claim them, at a rate that is hard to picture. The transactions doing it are peculiar to look at: a wallet sends coins to itself, attaches the claim, and pays the least the network will carry. One input, two outputs, about a hundred and forty bytes. The shape has a plain explanation. To carry a claim you need a transaction, and the cheapest transaction that is still valid is one that spends a coin straight back to the wallet it came from. Nothing is being moved. The payment is a wrapper for the claim.

Hold that against the growth in Bitcoin's traffic and the two are very nearly the same number. The chain carries about twelve and a half million more transactions a month than it did four years ago, and July's claims came to about twelve and a half million. Almost the whole increase is this one activity.

The arithmetic here is worth sitting with. July's claims cost about thirteen bitcoin in fees, some eight hundred and fifty thousand dollars at the price bitcoin was trading at when we measured. The tokens they produced were worth, at the going quote for the token, a little over three hundred thousand. The claiming is running at roughly three times the market value of what it claims, and each new claim makes every existing unit fractionally less scarce.

Whether the people doing it expect that quote to move, are farming some future reward that counts mints, or have simply never done the sum, the ledger does not say.

On the fifteenth of July, half of all Bitcoin transactions carried the same claim, byte for byte. Five wallets account for close to a third of everything the chain carried that month, and the commercial address-labelling service we checked can put a name to only one, a small exchange. That is less unusual than it sounds. We have followed an unlabelled wallet through three exchanges before, and the gap between what moves on a chain and what compliance tooling can name is wide.

The vote

A soft fork is a tightening of Bitcoin's rules, and it needs miners to go along with it. They show support by setting a flag in the blocks they mine. If enough blocks in a fixed window carry the flag, the rule locks in. BIP-110 asked for a temporary one-year limit on non-financial data, and set the bar at fifty-five percent of blocks.

Support was reported as reaching two and a half percent and failing there. Counting the flag ourselves gives a different shape.

Support had been climbing for eight straight windows, roughly doubling towards the end. The two and a half percent that was quoted is the top of that climb, not the level at which the proposal died. When the final window opened, the one where signalling was compulsory, it went to nothing. The last block ever to carry the flag was mined a little after five in the morning, universal time, on the eighth of August. The window opened fourteen and a half hours later. Not one block since has carried it, out of more than fourteen hundred.

The flag went from its highest reading ever to none at all inside a single day.

What the ban would have caught

BIP-110 had three clauses. Because they were written as precise byte limits rather than intentions, it is possible to run them backwards over the chain and see exactly what each would have removed.

The first clause was the real one, and it arrived about three years after the thing it was written for. The third is the oddest. In a release last October, Bitcoin's reference software raised an old limit on how much data a marker output could carry, from about eighty bytes to a size no ordinary transaction would reach. The number of such outputs has since risen more than fivefold, and almost none of that growth is in large ones. The flood is real and it is made entirely of things that were always allowed.

None of the three clauses would have touched the token claims that now occupy a third of every block, because the claims sit inside every limit it set. The marker they travel in is eighteen bytes long.

What the chain says

Bitcoin today runs close to the busiest it has ever been and collects less than one percent of miner income from it. It carries the same number of payments it carried four years ago. The people sending those payments briefly paid the highest prices in half a decade, for reasons that had nothing to do with them, and the permanent cost of that period is still being carried by everyone who runs the software.