Eighty bytes of forever

The first block of the Bitcoin chain, mined in January 2009, carries a line of text that has nothing to do with money. In the field where a miner is allowed to write anything, Satoshi Nakamoto put a newspaper headline: The Times 03/Jan/2009 Chancellor on brink of second bailout for banks. It works as a timestamp, since nobody could have written it earlier, and it works as an argument, since the headline is about the failure of the institution the chain proposes to replace. Both at once, in sixty-four characters.

A gesture with a very long history

That is not a new kind of writing. It is a foundation deposit. Mesopotamian kings buried inscribed cones in temple walls recording who built the thing and when; Roman builders stamped bricks with the consular year. The convention is that the first mark on a permanent structure states the date and makes a claim, and the two jobs are done by the same sentence because a date is only interesting if something is being asserted alongside it.

What is new is who else can write. A temple foundation deposit was commissioned. A brick stamp was institutional. The Bitcoin ledger accepts writing from anyone willing to pay the fee, and the result, over sixteen years, is the largest unsupervised body of inscription humans have ever produced.

The specimen

Here is what one of these looks like before anyone has read it.

Twenty-eight bytes. 65 is e, 6c is l, 20 is a space. Run the whole string through the table and it resolves to el sendero contigo es camino, which then has to be read a second time, in Spanish, to arrive at something like the path with you is a road.

There is no compression, no encryption and no cleverness in the first step. Someone typed a sentence, converted it to hex, attached it to a payment they were never going to get back, and paid a fee to have several thousand computers agree to keep it. The technical vocabulary around this material is forbidding, and the material itself is not. It is a person writing a line on a wall, using the most expensive wall ever built, and choosing to write it in a language most of the people passing will not read.

Every writing surface imposes terms. Clay wants wedges, because a stylus pressed into wet clay makes a wedge and does not easily make a curve. Stone wants straight strokes and rewards abbreviation, because every letter is an hour of a mason's day. The Bitcoin ledger wants eighty bytes, and it charges you for them.

How the space was won

The early method was crude. If you wanted to store data, you disguised it as a payment: you took your bytes, formatted them to look like a recipient address, and sent a tiny amount of currency to that non-existent destination. The message was preserved, because every full copy of the ledger keeps every transaction. But the fake addresses were preserved too, permanently, in the working set of unspent outputs that every node has to hold in memory. You were writing on the wall by driving a nail through the structural beam.

The developers noticed, and in March 2014 there was an argument. Jeff Garzik put the objection bluntly: using full nodes as dumb data storage terminals, he said, "is simply abusing an all-volunteer network resource." Third-party projects had "simply flipped an 'on' switch and started using bitcoin P2P nodes as unwanted data stores."

The compromise was a small piece of syntax called OP_RETURN. It marks an output as provably unspendable, which means nodes can recognise it as data and drop it from the working set instead of carrying it in memory forever. In exchange, the data is allowed. You sacrifice the money to buy the space, and the network stops arguing with you. Bitcoin Core 0.9.0 allowed forty bytes; the limit was later raised to eighty, where it sat for a decade, and it was still being fought over in 2025.

Eighty bytes, unevenly distributed

Eighty bytes is about a line of text, and this is where the surface stops being neutral between languages.

The encoding almost everyone uses is UTF-8, which spends a different number of bytes on different scripts. Unaccented Latin costs one byte per character. Cyrillic, Greek, Hebrew, Arabic, Persian and Georgian cost two. Bengali, Thai, Amharic, Devanagari, Chinese, Japanese and Korean cost three.

So a permanent sentence in English is twice the length of a permanent sentence in Russian and three times the length of one in Thai. For the Indic and Thai scripts it is worse than the table suggests, because a written syllable is often several code points, a consonant plus a vowel sign plus a mark, so the usable count drops below twenty-six.

The arithmetic runs the other way for Chinese. Twenty-six hanzi is a considerable amount of language; it is a classical quatrain with room left over. Bengali gets the byte penalty without the density that offsets it. Nobody designed this, and nobody has ever proposed fixing it, but there is now a permanent global inscription surface that charges by the byte and therefore taxes writing systems at three different rates.

The price of forever

On top of the byte limit there is the money. The transaction that carries the message pays a network fee, and the amount sent to the unspendable output is destroyed. Writing a sentence into the ledger has cost, over the years, somewhere between a few cents and tens of dollars depending on how congested the network was on the day. It has never been free.

That cost turns out to be the most interesting thing about the corpus. Data embedded this way has never accounted for more than a small fraction of Bitcoin's traffic. Given an unerasable surface and a real price, people did not fill it. They wrote one thing, carefully, and stopped.

Read enough of it and the categories sort themselves, and they are not the categories a records manager would predict. There is very little that anyone would file. There is a great deal that anyone would recognise.

Love, and its adjacent forms

The largest single category is people saying they love other people. I love Liwen and Wenxin forever. by Jun. I fell inLove withZhouYu at first sight. But I have the duty of my family & my boy. I really wanna tell her these in person. A marriage between Grishchuk Roman and Gres Anna is recorded in a transaction from April 2018, backdating the event to 2015. A proposal reads Angel will you marry me? -Alen, and sits four lines away from the Spanish line this essay opened with.

The register is consistent and it is not the register of the internet. It is the register of the headstone and the tree trunk. People writing on something permanent write shorter, plainer and more formally than they write anywhere else, and they say the thing they would be embarrassed to say at a normal length.

Memorial

In October 2020 someone recorded the death of Yang Yiping, 1957 to 2020, across a set of transactions in German, Chinese and Japanese. Three languages, chosen presumably because they were the three the man's life had been conducted in, on a surface that will hold them for as long as the network exists. It is a trilingual funerary inscription, which is a form with a genealogy running back through the Rosetta Stone to the Behistun relief, and it has been reinvented from scratch by a private individual who was almost certainly not thinking about Darius I.

Aphorism

NOTHING BIG CAME OUT OF SOMETHING SMALL. ...BUT SWEAT MEMORIES REMAIN. SMILE TO LIFE AND LIFE WILL SMILE BACK AT YOU. HE WHO IS GIVEN EVERYTHING HAS NOTHING. Time is always against us.

And, repeatedly, The Matrix: The truth is that there is no spoon, Welcome to the real world, Follow the white rabbit.

Given a permanent surface and one line to fill it, a substantial number of people quote something. This is also not new. The graffiti at Pompeii quote Virgil, badly and from memory, on walls where the writer could just as easily have written their own sentence. There seems to be a reliable human instinct that permanence calls for a borrowed voice, as though your own words are not formal enough for the occasion.

A country's founding document, in the wrong spelling

The longest human text in the corpus that I have been able to read is Indonesian. Running to something over a thousand characters, far past anything a single eighty-byte output could hold, it carries Sukarno's address of 17 August 1945 and the Proclamation of Independence itself: KAMI BANGSA INDONESIA DENGAN INI MENYATAKAN KEMERDEKAAN INDONESIA, closing with the dateline DJAKARTA, 17 AGUSTUS 1945 and the line Kita sekarang sudah bebas, we are free now. The material around it in the same dump dates from 2019.

Whoever put it there was making the obvious point: a founding document, placed somewhere no government can revise it. What they actually preserved is more interesting than that.

The Proclamation as typed in 1945 reads menjatakan, jang, tjara, sesingkat-singkatnja. Indonesian spelling was reformed in 1972, and the reform is precisely what changed those forms: j became y, dj became j, tj became c, nj became ny. The text in the ledger is in the post-1972 orthography throughout. It is not the 1945 document; it is a modern transcription of it, made permanent.

And it is not even a consistent one. Djakarta in the dateline was left in the old spelling, the way English keeps Ye Olde for effect, while every other word in the same passage was modernised. Somebody made an editorial decision, probably without noticing they were making one, and that decision is now as unremovable as the Proclamation.

This is the ordinary condition of transmitted texts and it is why textual criticism exists. What is unusual is the guarantee. A manuscript tradition gives you variant readings and lets a later editor weigh them. This gives you one reading, chosen by an anonymous person in 2019, and forecloses the argument permanently.

The accidental

Not everything in the corpus was written to be read. Among the messages are strings like EUR 18/Aug/16 SC1 0 BTC/100 USD PWD:XDRGIjssD#SFZWW7, which is a trading credential that somebody pasted into a field they had misunderstood. It is now permanent, worldwide, and unremovable.

Sitting between the deliberate inscriptions are the miners' own marks: Mined by AntPool, /ViaBTC/We Love Bitcoin!/, pool addresses, server version strings from misconfigured software. These are colophons. The scribe signs the manuscript, the mason cuts his mark into the underside of the block, the mining pool tags the coinbase field, and in every case the signature outlives the reason for it and becomes the thing historians use to date the work.

Advertising, inevitably

TD Ameritrade placed an advertisement in the ledger in April 2018, fragmented across multiple transactions and scrambled. A security conference sold tickets through it. A hardware wallet company marketed itself in it. Any surface that people read will eventually be sold to people who want to be read, and the interval between the invention of permanent public writing and its first commercial use appears to be short in every medium anyone has tried.

None of this material is indexed. The ledger has no table of contents, no metadata describing what any given payload contains, and no way of distinguishing a poem from a corrupted image header. Everything above was found the way a damaged tablet is read: by pulling fragments out of a bulk medium and working out which ones are language.

The tooling is exactly as crude as it sounds

The standard method is the Unix strings command, which walks through a binary file and prints any run of printable characters longer than some threshold. Anduck's BitcoinStrings, which publishes the results for the whole chain, uses strings -n 20, meaning runs of at least twenty characters. Ciro Santilli's inscription indexer does something more careful, tying strings back to the transactions that carry them.

The output has the texture of an epigraphic corpus. Word boundaries are unreliable. Fragments start mid-sentence because the twenty-character threshold clipped the front off. Random bytes from adjacent binary data attach themselves to either end of otherwise clean text, so a message arrives as (j&I love Liwen and Wenxin forever.by Jun. Languages are mixed on the same page with no marking. There is no attribution, no date on the object itself, and no context beyond position.

The numbering trap

The files on BitcoinStrings are named blk00192.txt, blk00604.txt and so on, and it is natural to read those as block numbers. They are not. They correspond to the raw storage files that Bitcoin Core writes to disk, each of which holds thousands of blocks. Block 192 was mined in January 2009, five years before OP_RETURN existed; the messages in blk00192.txt are from a completely different period.

That is a small error and worth naming, because it is the characteristic error of reading an unfamiliar archive: the numbering of the container mistaken for the numbering of the contents. Anyone who has worked with a box list that does not match the folder list has made the same mistake.

Counting it

The one serious quantitative survey remains Matzutt and colleagues at RWTH Aachen, who in 2018 went through the chain systematically and found more than sixteen hundred embedded files alongside a very much larger volume of short text. Massimo Bartoletti and Livio Pompianu had already classified the OP_RETURN metadata by protocol and purpose in 2017. Between them they establish the shape of the corpus, and both are now years out of date, because the arrival of Ordinals in 2023 changed the economics of writing to the chain and nobody has recounted since.

It is tempting to call the result an archive. On reflection the word does not fit, and the ways it does not fit are the interesting part.

Everything an archive does, this does not do

An archive appraises: somebody decides what is worth keeping and what is not, and the decision is the profession. This keeps everything, because it has no mechanism for not keeping things. An archive arranges and describes, so that a thing kept can later be found. This has no arrangement beyond chronology and no description at all. An archive maintains provenance, the record of who made a thing and how it came to be held. Here the provenance is a transaction identifier and a wallet address, which tells you nothing about a person. An archive can deaccession. This cannot.

What is left when you subtract all of that is not an archive. It is a deposit. The nearest real comparison is the one this section keeps returning to: the tablet archives of Mesopotamia survive in bulk not because anyone curated them but because the buildings holding them burned down, and fire hardens clay. Most of what we have is receipts. The literature and the letters that people actually read now are a thin seam running through several hundred thousand accounting documents, and they were preserved by exactly the same accident that preserved the accounts.

The Bitcoin ledger will present a future reader with the same ratio and the same problem. Overwhelmingly transactional, with human writing scattered through it, and no signal marking which is which.

Permanence is not preservation

Alexandra Woodall and Sharon Ringel made this argument about blockchain preservation generally: the appeal of the technology to the archival imagination rests on a conflation of durability with trustworthiness, and the two are not the same property. Victoria Lemieux, who has spent longer than most testing the claim, reached a similar place from the recordkeeping side.

The bytes will survive. Whether anyone can read them is a separate question with a separate answer, and it depends on things the chain does not guarantee: that the character encoding is still understood, that the languages still have readers, that somebody has done the indexing work, that the context making a sentence legible has been recorded somewhere the sentence is not. el sendero contigo es camino is durable. It is not self-explanatory, and it never will be.

The closing thought

The governance conversation about this technology is almost entirely about deletion; whether a right to be forgotten can survive a system with no delete function, and what a retention schedule means on a surface that does not forget. That conversation is real and it is not going away.

But it rests on an assumption worth questioning, which is that remembering and not-deleting are the same act. The chain will hold el sendero contigo es camino for as long as the network runs. It does not know who wrote it, or who they were walking with, or whether they are still walking, or what the line was answering. It does not know that the Indonesian passage a few thousand lines away is in the wrong century's spelling. It has kept the bytes and lost the meaning, which is the ordinary condition of every inscription humans have ever cut into anything, and the reason the discipline of reading them exists at all.

Deletion is impossible here. Forgetting is proceeding normally.