Ordinals, and why the numbering is a convention

An inscription is the second half of the idea: a way of attaching arbitrary data to a specific satoshi so that the data travels with it when it is spent. The mechanism uses the witness field that SegWit introduced and that taproot extended. A taproot spend can carry a witness script and its inputs, and the witness is not covered by the same block-weight accounting as the rest of the transaction, which is what makes it a practical place to put a payload.

The construction is a commit-and-reveal pair. The first transaction commits to a taproot output whose script tree contains the data, typically by embedding a hash of the payload in a script that the spender can later satisfy. The second transaction spends that output and reveals the payload in the witness, where an indexer can read it. The Ordinals documentation describes the envelope format used to mark the payload — a sequence of opcodes that delimits the content type and the body — and the reference implementation is what defines the exact rules for parsing it.

Because the payload lives in the witness, it is subject to the witness discount rather than the full block weight, and it is prunable in the sense that a node that does not need historical witness data is not required to keep it. That is a real property of the design and it is also the source of the argument about it: the data is cheap relative to its size, it is stored by nodes that keep the full chain, and it is not something the protocol was designed to carry. None of that makes it invalid. It makes it a use of the protocol's own rules that the protocol does not distinguish from any other spend.

Everything that makes ordinals useful — the numbering, the attachment, the notion that a particular satoshi carries a particular inscription — is produced by software reading the chain, not by the chain itself. A Bitcoin node validates the transactions and stores the blocks. It does not know that satoshi number 1,234,567 carries a text file, and it would not care if it did. The meaning is supplied by the indexer, and the indexer is a separate program with its own rules.

This has consequences that are easy to miss. The first is that the numbering depends on a consistent reading of the chain, including blocks that were later reorganised away: the ordering rules have to say what happens to satoshis whose first appearance was in a block that did not survive, and the answer is a convention rather than a fact. The second is that inscriptions are only as durable as the indexers that recognise them. A satoshi can be spent, merged into a larger output and split again, and the tracking rules decide where the inscription is deemed to have gone. The third is that the ecosystem's agreement is what gives the convention its value; a single indexer with different rules would produce a different, and equally unfalsifiable, history.

The comparison with the unspent output model is the clearest way to hold the distinction. The UTXO model is consensus: every node agrees on which outputs exist and what conditions spend them, and a transaction that violates those conditions is rejected everywhere. Ordinal numbering is not consensus: it is a layer of interpretation applied on top of a ledger that is deliberately indifferent to it.

An inscription occupies block space, and block space is the scarcest resource the protocol produces. The fee for an inscription is therefore set by the same market that prices every other transaction: the sender bids for inclusion, and the bid has to compete with payments, consolidations and every other spend waiting in the mempool. A large inscription is a large transaction, and a large transaction that wants prompt confirmation has to pay accordingly.

The cost is not only the fee. Every full node that keeps the chain stores the witness data, and every node that relays transactions carries it across the network. The witness discount means the marginal cost to the sender is lower than the marginal cost to the network in storage terms, and that gap is the substance of the long-running objection to the practice. It is a genuine trade-off rather than a technicality: the protocol's fee market allocates space efficiently among competing spends, but it does not price the long-term storage externality, and inscriptions are the clearest example of a use that exploits the difference.

The counter-argument is equally worth stating. The rules that make inscriptions cheap are the same rules that make taproot spends cheap, and those rules were adopted deliberately to improve privacy and reduce the cost of complex scripts. A protocol that permits arbitrary witness data cannot then complain when someone supplies it. The honest summary is that inscriptions are a legitimate use of the protocol's rules with a real cost that the fee market does not fully capture, and that reasonable people weigh those two facts differently.