# The βook of βitcoin — a brief > The whole book explained in one file, for a reader who has this document and > nothing else: a person new to it, or an assistant asked about it without the > means to fetch eight URLs. Everything below is checkable against the book at > and the chain itself. ## What it is A verbatim translation of the Bitcoin block chain into human language. Every byte of every transaction is carried in the words of grammatically correct prose, and decodes back out exactly — filter a passage against the payload wordlist and the transaction returns, byte for byte, in order, with nothing added and nothing omitted. The prose is generated, not written. A Rust engine called Glossia () encodes the opaque bytes — hashes, keys, scripts, witness pushes — as sentences, invertibly; the book is set in Latin by default and reads in English, Čeština or Deutsch on request. The words differ, the record does not. No sentence of the book's own prose is composed by a person. The argument this exists to make: **a transaction is a form of speech.** It can be set down in prose, read aloud, quoted, transcribed by hand, and carried back to the chain intact. That is a demonstration rather than a metaphor, and every page is a witness to it. ## How it is arranged The divisions were already in the chain; the book only gave them the names a book gives its parts. | Part | Is | Mark | |---|---|---| | Volume | a halving era — 210,000 blocks | Roman numeral | | Book | a difficulty window — 2,016 blocks | β | | Chapter | one block | ■ | | Section | one transaction, 1-based within the block | § | | Page (folio) | the chain-wide running transaction count | a bare number | So a passage cites itself as `III β2 ■5 §1` — the 1st transaction of the block at that address. An output appends its 0-based vout (`§1.0`); a witness appends its footnote letter (`§1.a`). The arithmetic, from a block height `h`: ``` volume = ⌊h / 210000⌋ + 1 book = ⌊(h mod 210000) / 2016⌋ + 1 chapter = (h mod 210000) mod 2016 + 1 height = (volume−1)×210000 + (book−1)×2016 + (chapter−1) ``` **The one place a name stops being exactly true**, and the book leaves it visible rather than papering over it: book numbering restarts with each volume, and a halving era is not a whole number of difficulty windows (210,000 blocks is about 104⅙ of them). So every volume's last book is a short one — 336 blocks against 2,016 — and from Volume II on, an era begins 336 blocks into a window already running, a further sixth of a book late each time. A book after Volume I is still 2,016 blocks but is not *the* 2,016 blocks the network weighed: it carries the tail of one window and the head of the next, with a retarget inside it. They come back into step at Volume VII, six eras being exactly 625 windows. Supply and difficulty are two different clocks, and they agree once every six eras. ## What the pages look like - **A volume's title leaf** — the numeral, the books it spans, the years, and the work the era cost, summed from the headers: `+6.29·10²⁰ hashes` for Volume I. Each volume so far has cost more than every volume before it put together. - **A book's title leaf** — the target its chapters were mined under, written as the inequality it is, over the chapters it holds and the days it took. - **A chapter's title page** — the block hash said in prose between `⌘ᵐ` and `⓪ⁿ` (the two counts always summing to 256), then the eighty header bytes in wire order: version, previous block, merkle root, timestamp, target, nonce. - **A section** — one transaction, opening on its § number, any name it has been given, its transaction id in prose, the time it was recorded, and the confirmations over it. A section is set as a manuscript page in three columns: - the **left margin is provenance** — the citation of the output each input spends, with that input's sequence mark - the **body is the prose**, in wire order: version, inputs, outputs, locktime. An output's script is titled by the term it binds, and the locktime closes the page centred like a colophon - the **right margin is consequence** — the output's amount and, beneath it, the citation of the section that later spent it. **An output with no citation there has not been spent: that absence is the UTXO.** Beneath them a balance line states the sums: inputs at the left, outputs at the right, the fee between, signed negative — no labels, the geometry is the equation. It is derived arithmetic rather than bytes on the wire. A coinbase inverts it: the subsidy at the left, the block's gathered fees between as `+ fee`, since the miner's take arrives where every other section's leaves. **Witness data gathers at the foot as footnotes**, lettered a, b, c… through the whole alphabet, continuing in bijective base-26 — aa after z. It sits mid-serialization on the wire, but it is spending apparatus rather than speech, so it goes to the foot and takes its input's citation and amount with it. Even its type size is consensus's rule: a witness byte weighs a quarter of any other, and the footnotes are set to that same discount. ## The notation Scripts are printed in an alphabet of marks, one per opcode — `⧉` OP_DUP, `∇` OP_CHECKSIG, `⌖` OP_HASH160 — and the pushes between them as prose. A mark's superscript counts the **bytes** of the push beside it (`h²⁰`, `p³²`), which is what keeps the form invertible. `⌘` alone counts bits, so that it pairs with `⓪`'s proof-of-work zeros to 256. The full table — all 110 opcodes, the citation sigla, the term notation, and the marks a passage carries — is at . Four marks are the book **declining rather than guessing**, and they matter when reading a page: `∅` nothing is there, `⋯` not yet known, `☒` two readings that disagree, `…` a value the engine could not say. A value that cannot be computed honestly is never filled in with a plausible one. ## Scripts as terms Every locking script is an abstraction over its own datum, so the book titles each output with the term it binds — the kind, `:=`, then the lambda: ``` P2PK := λp. p ∇ P2PKH := λh. ⧉ ⌖ h ≡ ∇ P2SH := λh. ⌖ h = P2WPKH := λh. ⓪ h P2WSH := λh. ⓪ h P2TR := λp. ① p ``` The term is **read, not looked up**: every push becomes a binder, every opcode stays where it stands. So a script with no tabled form is titled too, and one that binds nothing is not titled at all. A closed normal form settles on chain by itself. The free variables of an open one name exactly what a spender must still supply — a signature `s`, a key `p`, a redeem script `r`, a witness script `w`, a tapscript leaf `t`, a control block `c` — which is the typed interface to the spend. A script-hash output commits only to a hash until it is spent, so what it demands is written `( r )` and the binders it cannot count are written `…`; the spending transaction is where the abstract becomes concrete. An address, in this reading, is not the script but the **redex that reduces to it**: `(λp. ① p) p³²` is an application, and performing it gives the scriptPubKey byte for byte. ## The record and the readings of it This is the distinction to preserve when quoting the book, and the one most easily lost. **The record** is what the chain says — this output moved to this script at this height. It has no author. A passage *is* its transaction, byte for byte, and anyone can check it against the chain. **The readings** are what people say *about* the record: that a block is worth noticing, that a transaction bought a pizza, that an address belongs to a named party. Every one is a claim, made by someone, resting on evidence that can be examined and can be wrong. The book keeps them in different registers, visibly — a reading is set apart, credited to whoever wrote it, and licensed separately. So: quote the prose as the chain's speech, and attribute commentary as somebody's reading, by name. "Ten thousand coins moved to a script" is the record; "they bought two pizzas" is testimony from the people involved. Do not fold the second into the first. Titles like *Bitcoin Pizza Day* are editorial, not chain data. ## The appendices The back matter holds what reading order cannot carry, reading order being the order blocks were mined. - **Appendix I · The Mempool** — the chapters not yet bound: the next chapter as itself, and the same transactions ranked by amount and by size. - **Appendix II · The Mines** — pools by their share of the last difficulty window, each share printed with its own standard error. - **Appendix III · Consensus** — soft forks under their BIPs, each with its ballot replayed block by block, including the forks the chain declined. - **Appendix IV · The Ledgers** — an address is a name rather than a place, so a ledger is cited by nothing and discovered instead. - **Appendix V · Inscriptions** — envelopes read out of witnesses, with what is convention rather than consensus marked as convention. - **Citations** — works outside the text, dated into a chapter by their OpenTimestamps proofs, standing last and unnumbered. ## Where the text actually lives The pages compose their prose in the browser, so fetching an app URL returns a shell with no passage text in it. The text is at: - — the curated passages, each pre-rendered as plain markdown, commentary last under its own heading - — the notation key - — the book in its own voice, at length - — the URL grammar, and how to reconstruct any passage on the chain with the published engine - `/III/2/5/1/` — a curated passage at its citation, as an ordinary HTML page Anything not curated is not pre-rendered; it is composed on demand in the browser, or reconstructed with the engine as `/llms.txt` describes. ## What to avoid saying about it - That a translator chose the words. Nobody wrote the prose; an engine encodes the bytes, and the same transaction can be rendered in more than one way. - That a passage is a summary or a paraphrase. It is the transaction, and it decodes back to it. Fidelity wins over fluency every time. - Quoting a curated title or a commentary note as though the chain said it. - Inventing a passage's text. If the prose for a given transaction is not in hand, say so: the engine is deterministic and can be run, but its output cannot be guessed, and a plausible sentence is a fabricated one. ## Licensing The prose (the chain's own speech) and the notation: CC0 1.0, public domain. The curated contents, names and annotations: CC BY 4.0, credited. Commentary by others: theirs. The machinery: MIT OR Apache-2.0. This brief is part of the editorial layer, and is CC BY 4.0.