What the report actually looks like.
A nine-hour novel, eleven tracks, proofed in four minutes. This is the whole report. Not a summary of it, not the first page of it. The whole thing.
A worked example on an invented title. We do not publish output from real productions, which is rather the point of how we handle your material.
Proof report: The Quiet Yard
9 amendments · 2 pronunciations to confirm
| Ch 2 | 14:22 | Word missed. Book says "she had never once asked", audio says "she had never asked". |
| Ch 2 | 31:08 | Wrong word. Book says "death", audio says "debt". |
| Ch 3 | 08:47 | Incomplete word. Book says "Marchetti", audio says "Marche." |
| Ch 4 | 22:15 | Sentence missed. Book says "The letter arrived on a Tuesday, which he would remember for the rest of his life." |
| Ch 5 | 02:31 | Wrong word. Book says "melded", audio says "molded". |
| Ch 5 | 47:56 | Words added. Audio says "and then he simply walked away", book says "and he walked away". |
| Ch 7 | 11:04 | Word missed. Book says "almost every morning", audio says "every morning". |
| Ch 8 | 39:19 | Wrong word. Book says "affect", audio says "effect". |
| Ch 9 | 05:52 | Line repeated. The sentence beginning "By then the yard was empty" is read twice. |
Pronunciation to confirm
| ch 3, 08:47 | “Marchetti” is read as “Mar-CHET-ee”, 31 times throughout |
| ch 4, 19:02 | “Aoife” is read as “AY-oh-fee”, 12 times throughout |
Each name is listed once. A name read the same way throughout is usually a choice to confirm with the publisher; a name read differently only once is worth a listen.
Why it is nine lines and not four hundred
The engine found several hundred differences between the audio and the manuscript. Almost none of them were faults. A proofer who is handed four hundred lines checks the first thirty and stops trusting the tool, so the job is not finding differences. The job is knowing which ones a person should spend a minute on.
These are real suppressions from a real run:
| Difference found | Why it is not in the report |
|---|---|
| "1,847" read as "eighteen forty-seven" | Correct reading of a numeral. Never a finding. |
| "colour" in the book, "color" heard | Identical sound. A spelling difference cannot be an audio fault. |
| "THE UNION OF THEIR DREAMS 47" mid-paragraph | A running head and page number the PDF left behind. |
| A 340-word block absent from the audio | An index, a photo caption and a copyright page. Not narrated, correctly. |
| "their" in the book, "there" heard | Homophone. The narrator said the right thing. |
Nothing is thrown away. Everything the engine found is in your dashboard behind the report, with the audio on a play button, so you can accept a finding into the report or decline it out of it. Declined lines are struck through rather than deleted, in case you change your mind.
The publisher sheet
Most studios want the same five columns, so that is what exports. File number, timestamp, chapter, page in the printed book, and the problem in a sentence. Page numbers come from your actual PDF, not an estimate.
| File | Time | Chapter | Page | Problem |
|---|---|---|---|---|
| 03 | 00:31:08 | 2 | 44 | Reads “debt” for “death” |
| 05 | 00:22:15 | 4 | 91 | Sentence omitted, “The letter arrived on a Tuesday…” |
| 10 | 00:05:52 | 9 | 203 | Line read twice, “By then the yard was empty…” |
It also exports as a printable report and a CSV, and it fits the .odt template most houses already circulate.
What it will not do
It will not catch everything. No tool will, and anyone selling you one that does has never proofed a book. It hears every word of the audio against every word of the manuscript, which is the part a person does badly at hour nine, and it hands you a short list. Deciding what is actually wrong is still your job. That is the deal, and we would rather say so here than in a footnote.