Fable
The review room fixes the edges first; the senior editor only looks at what truly changed
The bookshop in the city revised contracts for the chamber of commerce. The senior editor was familiar with the clauses and could spot subtle risks, but each time he edited word by word from the first page to the last.
As contracts multiplied, many changes were just dates, titles, and boilerplate disclaimers. The senior editor still personally went through every page, and the genuinely complex liability clauses ended up at the back.
The manager first asked the client to make fewer changes, then required editors to use templates. Templates save a little, but as long as source and target text are interleaved, major edits still require slowly confirming every position. Someone had junior editors submit finished drafts directly. As a result, they changed one word wrongly in a key liability paragraph, and a sum had to be paid out afterward.
Later the review room changed to two steps. A junior editor first proposes a "possibly complete" revision based on common patterns, marking which positions are merely mechanical replacements and which may need an editor's judgment.
The senior editor no longer rewrites all the text, but verifies whether these changes can be accepted; passages that are accepted consecutively pass directly, and only places with disagreement return to word-by-word editing.
This approach is especially suitable for "old draft to new draft" tasks. A large amount of text will be retained anyway, and the system only needs to intelligently guess which fragments can be reused and which need to be changed.
The shopkeeper understood that editing-type generation is different from writing from scratch. Its opportunity lies in using existing text, letting a fast assistant propose changes, and then having a strong editor verify them.
Later, when the scriptorium received code patches, contract revisions, and customer service rewrites, it would first assess the proportion of reusable original text. The more stable the original text, the more valuable speculative editing became. Later the scriptorium applied this method to a contract with major changes, and the results were worse. Junior editors guessed too much, senior editors kept sending it back, and it took longer than reviewing from scratch. The manager then added a criterion: only when the old draft and the new goal are largely similar should the assistant guess first; if the structure has completely changed, hand it directly to a senior editor for planning. The value of speculation comes from reusable original text, not from blindly starting early. Later the senior editor required recording the acceptance rate every time. If only three out of ten guesses are correct, the assistant should not run first; if eight can be accepted, it is worth continuing. Accurate guessing is what saves effort; inaccurate guessing just adds another round of rework. Assess similarity first, then talk about speed.