What happened
On 20 January 2023 Casey Rodarmor announced that inscriptions were ready for Bitcoin mainnet, alongside the release of ord 0.4.0. An inscription attaches content, an image, a text file, anything at all, to an individual satoshi, and stores it inside a Bitcoin transaction rather than pointing at a server.
It stores it in a Taproot script. The Ordinal Theory Handbook is direct about where the bytes live: content is "entirely on-chain, stored in taproot script-path spend scripts."
The script is written so that it never executes:
OP_FALSE
OP_IF
OP_PUSH "ord"
...
OP_ENDIF
OP_FALSE OP_IF opens a branch that is never taken, so everything up to OP_ENDIF is
carried by the transaction and ignored by the interpreter. Because a script-path spend needs
an existing Taproot output, an inscription takes two transactions: a commit, then a reveal.
What it changes
It changes what data costs, and it does so without changing a single consensus rule.
BIP 141 defines block weight as "Base size * 3 + Total size", capped at four million. Read the formula slowly. A byte outside the witness is counted by both terms, so it costs four weight units. A byte inside the witness is counted once, so it costs one. Witness data is charged a quarter of the rate charged to everything else, and inscription content is witness data.
The second piece is BIP 342, which sets the rules for scripts under Taproot: "The maximum script size of 10000 bytes does not apply. Their size is only implicitly bounded by the block weight limit." Before Taproot a script could not hold much of anything. Now the only ceiling is the block.
One older limit survives: BIP 342 keeps the 520 byte cap per stack element, so content is chopped into 520 byte pushes.
Neither rule was written with this in mind. One prices data a node can discard; the other lets a spending condition be arbitrarily elaborate. Together they mean a megabyte of image occupies the block space of 250 kilobytes of payments.
What it does not change
Bitcoin does not know what an inscription is. A node sees a valid Taproot spend with a large witness, validates it, and moves on. There is no new opcode, no new output type, and nothing to activate.
Bitcoin also does not track individual satoshis. Ordinal theory numbers them by convention,
in mining order and then first in, first out through a transaction's inputs and outputs.
That convention lives in the ord indexer, not in consensus, so two indexers that disagree
produce two answers and the network has no opinion on which is right.
The block space limit is untouched. Inscriptions create no room. They bid for the room that exists, against everyone else who wants it.
Context
The witness discount is not an oversight and not new. It arrived with SegWit in 2017, which split signature data out of a transaction and priced it lower, since a node can discard it once checked.
Taproot activated in November 2021 with an unrelated goal: letting a complicated spending arrangement look like a simple one. The tree of alternative conditions it introduced was designed to hold recovery paths and co-signer requirements. It holds whatever fits.
This has been running for weeks already. Rodarmor made the first mainnet inscription on 14 December 2022, in block 767,430: a 793 byte PNG. The command that creates one stayed disabled for mainnet until 9 January. What happened on 20 January was the announcement.
Arbitrary data has been in the chain since the genesis block, whose coinbase carries a newspaper headline. What is different is the price and the volume, not the possibility.