Ordinals and inscriptions: how images and text ended up inside Bitcoin blocks

Ordinal theory numbers satoshis in the order they were mined

A satoshi is the smallest unit of bitcoin, one hundred-millionth of a coin. Bitcoin itself does not tell satoshis apart; a transaction output holds an amount, not a bag of numbered coins. Ordinal theory adds the numbering from outside. It gives every satoshi a serial number in the order it was mined, and when a transaction spends coins it treats satoshis as moving from inputs to outputs first-in, first-out. With that one rule, any satoshi can be followed from the block that created it to wherever it sits today.

The convention lives entirely in software that reads the chain, chiefly a program called ord written by Casey Rodarmor. The documentation is explicit that ordinal theory “does not require a separate token, another blockchain, or any changes to Bitcoin.” Nodes do not know about it, miners do not enforce it, and two people running different software could in principle disagree about which satoshi is which. The docs also grade satoshis by rarity: the first satoshi of each block is “uncommon,” the first after each difficulty adjustment “rare,” the first after each halving “epic,” and the very first satoshi of the genesis block “mythic.”

An inscription is a file parked in a Taproot transaction’s witness

An inscription is a file attached to one of those numbered satoshis. The file is not stored in a transaction’s outputs, which hold amounts and spending conditions, but in the witness, the part of a transaction that carries signatures and scripts. This became practical with Taproot, the upgrade specified in BIP341 by Pieter Wuille, Jonas Nick and Anthony Towns and activated at block 709,632. Taproot lets an output be spent by revealing a script that was committed to in advance; in a script-path spend the witness carries the script itself plus a control block proving it belongs to the output.

The Ordinals documentation describes how inscription content rides inside such a script. The data is wrapped in an envelope, OP_FALSE OP_IF … OP_ENDIF, a construction the script interpreter treats as doing nothing, so the data is on the chain but never executed. A content type, such as an image or text format, goes first, then the bytes, in pushes of at most 520 bytes each. Creating one takes two transactions: a commit transaction that creates a Taproot output committing to the script, and a reveal transaction that spends it and puts the content on the chain.

Two rules make large files possible. BIP342, the companion tapscript specification, states that “the maximum script size of 10000 bytes does not apply” to Taproot scripts; their size is bounded only by the block weight limit. And witness data has counted at a discount toward block size since SegWit, which the Ordinals docs note makes “inscription content storage relatively economical.” Neither rule was written with inscriptions in mind. Both were on the books when someone found a use for them.

A trickle in December 2022, then a rush that filled every block

Rodarmor made the first inscription on the main Bitcoin network on 14 December 2022. On 20 January 2023 he published a post titled “Inscribing Mainnet” announcing version 0.4.0 of ord: “Inscriptions are finally ready for Bitcoin mainnet.” He described inscriptions as “digital artifacts native to the Bitcoin blockchain,” and set them against typical NFTs, tokens on other chains that often point to files stored elsewhere. Inscriptions, he wrote, are “immutable and on-chain, on the oldest, most decentralized, most secure blockchain in the world.”

His own account of what followed is short: “There was a trickle of inscriptions, then more, and then a rush, filling every block.” Within ten days CoinDesk was reporting that the Bitcoin community had erupted in “existential debate” over the project, with developers and commentators lining up on both sides. The argument that started that week has not really ended.

What it did to fees and block space in 2023

Block space is fixed, so anything new that wants in has to outbid what was already there. Our guide to fees and the mempool explains the auction; the short version is that miners fill each block with the transactions paying the most per unit of space, and everyone else waits. In early 2023 the new demand was mostly images. By spring it was mostly text, because a scheme called BRC-20 used tiny text inscriptions to represent tokens that could be minted and traded, and minting them meant sending a great many transactions.

The effect peaked in the first week of May 2023. CoinDesk reported nearly 400,000 unconfirmed transactions waiting, more than at the height of the 2018 or 2021 booms, and an average fee just over $8, double the level in March. It counted about 14,000 BRC-20 tokens with a combined market value of $482 million. On 7 May the exchange Binance paused bitcoin withdrawals, citing the congestion, before resuming within two hours. Colin Harper of Luxor Technologies told CoinDesk the difference from earlier fee spikes was that the “BRC-20 standard is a new way to inscribe.”

For miners this was welcome income. For anyone trying to make an ordinary payment, it was an expensive month. Both facts are true at once, and they are the real substance of the argument that followed.

Spam or a legitimate use? The debate has no referee, by design

The critics’ case is about cost and purpose. Luke Dashjr, a long-time Bitcoin Core developer, called Ordinals “an attack on Bitcoin.” Adam Back of Blockstream suggested miners might refuse the transactions as a discouragement. The concern is that files compete with payments for a scarce resource, raise fees for everyone, and add permanent bulk that every full node, meaning every computer keeping a complete copy of the chain, must store.

The supporters’ case is about rules and revenue. Peter Todd, another Core developer, pointed out that embedding data in transactions had always been possible; this site’s wallet library has an example from 2013, when the Bitcoin white paper itself was written into the chain. Dan Held argued it was “good for Bitcoin.” Supporters also note that the block subsidy shrinks on a schedule and fees must eventually pay for security, so new demand for block space is not obviously bad. Rodarmor’s own position was that the fee market is the referee: whoever pays most gets in.

The question reached Bitcoin Core’s repository on 5 September 2023, when Dashjr proposed extending an existing relay setting, which limits how much arbitrary data a node will pass along, to cover data hidden in witness scripts. Several maintainers objected to the concept. Peter Todd wrote that miners were unlikely to give up the revenue and that filtering “would simply encourage the development of private mempools.” The maintainer posting as glozow wrote that “we have no way of detecting whether a bunch of bytes are legitimate” and that deciding which uses are legitimate is not protocol development’s place. Another contributor argued the filter would be self-defeating, since inscribers would route around filtering nodes. The proposal was closed on 5 January 2024 without consensus.

That is where things stand. Nothing in Bitcoin’s consensus rules forbids inscriptions, no rule change to forbid them has come close to agreement, and node operators remain free to filter what they relay. Whether that is a healthy equilibrium or an unresolved problem depends on what you think Bitcoin is for, which is a question the software was deliberately built not to answer.