2 Copy
The hand compositor did not work from the author’s intentions. He
worked from copy: the manuscript, read and marked for the shop, cut into
takes by the overseer, and hung on a hook where the man could reach it.
The standing instruction in every shop was follow copy,
and the
old hands stretched it into a joke: follow copy though it fly out of the
window. A shop where each compositor improved the text as he set it
would print a book that nobody wrote.
So there was one copy, and it was law. In the reading closet the copyholder read the manuscript aloud while the reader followed the proof, and where the two differed, the copy won. When an author changed his mind on a proof, the shop charged him for the alteration; he had departed from copy. The marked proof then became the copy for the next pass. There was never more than one authority, and everyone knew where it hung.
Two authorities agree on the day you create the second one and on no day after. Someone corrects the proof and forgets the manuscript, or corrects one take and not its twin, and from then on the shop holds two versions and can prove neither.
2.1 Nothing stated twice
My shop keeps the same law with different furniture. The copy is a folder of chapter files. Everything a reader ever touches, the print pages, the screen edition, the pages on the web, is set from those files by the frame, the standing build that composes every format from the one manuscript.
The law reads: never state a fact a second time that the frame could derive. The title of this book appears in four places: the cover, the half title, the running heads, and the notice of each edition. It is written once, in a small file the frame consults, and set everywhere else by machinery. Write it by hand in a second place and the two agree for a while. Then a retitle touches one and misses the other, and the shop has two titles.
Derivation makes the disagreement impossible instead of unlikely.
2.2 The shelf orders itself
The reading order of the chapters is recorded nowhere. Each chapter file carries a number in its name; the frame sorts the names, and the sorted list is the book. There is no contents page kept by hand, no manifest.
A manifest would be a second copy of the shelf. It agrees on the day it is written. Then someone renames a chapter file, or slides a new chapter between two old neighbors, and the manifest goes on swearing to an order that no longer exists. The shelf cannot be wrong about its own order. When you want a fact, take it from the thing that owns it.
2.3 An index that cannot lie
The index is mined from the text on every build. I pick the terms worth finding; the frame reads the whole manuscript and locates their pages. The index cannot disagree with the book, because it has never seen anything except the book as it stands this minute. An index kept by hand starts disagreeing at the first revision and never stops.
The harder case is a curated term that matches nothing. The comfortable design would skip it quietly and print the entries that still work. This shop fails the build instead, and the failure names the term. A dead entry does not stay hidden. One reader follows it to a page that says nothing of the kind, and after that he doubts the rest of the column. So when a revision wears away the phrase a term depended on, the build stops and makes me choose: restore the phrase or retire the term. Either answer is fine.
2.4 The ledger is the notice
The shop keeps a ledger of editions. Every change goes in as it is made, one plain line under the heading of the edition to come: what was added, what was fixed, what was removed. When the edition ships, that section of the ledger becomes the release notice, word for word. Nobody sits down to write the notice. It was written all along, in the tense of work being done, by whoever did the work.
The notice cannot flatter, because nobody composes it. A summary
drafted at release time remembers the additions and forgets the
reversals. A ledger line written the hour a defect was fixed says
fixed
and names the defect. Copy the ledger in verbatim and the
customer reads what the shop recorded for itself.
2.5 Generator, checker, convention, memory
The devil is what makes this law urgent. Ask it to add a chapter, and if the shop keeps a hand-made contents page, a hand-kept index, and a drafted notice, it will cheerfully update all three, each a little differently, each version plausible on its own. It copies eagerly and never quite the same way twice. You will not catch this by reading harder. Leave it nothing to copy. Where a fact lives in one place, the devil can only get it wrong in one place, and one wrong place is what the frame is built to catch.
When a fact truly cannot be derived, rank the ways of keeping it: prefer a generator to a checker, a checker to a convention, and a convention to a memory. A generator produces the artifact, so the artifact cannot disagree with its source. A checker complains only after the disagreement exists, and only about disagreements someone thought to check for. A convention is followed until somebody has a busy Tuesday, and it fails without a sound. Memory is worse. Every step down the ladder trades a machine for a person paying attention.
The old shops ran the whole of this with one hook and one standing instruction. There was one copy, and it was dear. Everything else was proof, and proof was cheap: pull it fresh, mark it up, throw it away. The copy still hung where it always hung, and the next pull would be set from nothing else.