We measured the voice of our own novel

Every writer knows the feeling that a book gets tired somewhere in its second half. Before revising Dreamscape Code we tried to find out whether that feeling could be measured, so we ran the whole manuscript through a set of counts, chapter by chapter, and compared the opening third with the end.

What stayed the same

Rhythm, as it turned out. Mean sentence length ran at 10.7 words in the first eight chapters and 10.5 in the last four. Short sentences of five words or fewer held near a third of the total throughout. Whatever was going wrong late in the book, it was not the sentences getting longer or the pace slackening.

What changed

Two things, both sharply. Stock phrasing rose from about ten instances per ten thousand words in the early chapters to fifty-seven per ten thousand by the end: hearts pounding, jaws tightening, resolve hardening, the weight of things. And the narration stopped asking questions. Early on the protagonist wonders about everything, three times per thousand words. By the final chapters that has fallen to one every two thousand, and the book has started telling the reader what she feels instead.

Put plainly: the prose stopped noticing and started summarising. That is what the tired feeling in a last act usually is.

What we did with the numbers

They became a target. New and rewritten passages were measured against the early chapters and sent back if they fell outside the band: stock phrases under about ten per ten thousand words, narrated questions around three per thousand, dialogue tags dominated by said, asked and replied.

None of this decides whether a sentence is any good. A count cannot tell you that. What it can do is catch the drift early, and turn an argument about feel into a question with an answer.

We now run the same measurements on every manuscript we edit, and the author gets the numbers with the notes.