Why your proof and your production copies differ
You checked the proof, you approved the book, and months later you hold a copy a reader sent you to sign. The cover is a shade lighter than yours. The paper looks a little warmer. The fore-edge margin is not quite the same width.
Nothing changed. You did not upload a new file, and nobody edited it for you. Two copies of the same book, printed weeks apart in different buildings, are simply not required to be identical — and knowing which differences are normal is what stops you from re-uploading a file that was never wrong.
A proof is a sample, not a contract
A proof answers three questions, and it answers them well. Is the file right — does the spine land between the folds, does the artwork survive the trim, do the margins clear the binding? Is the content right — the title, the running heads, the chapter that opens on the wrong side? And does the object work in the hand at all?
That is the whole job, and it is worth doing properly: Order the proof. Then read it in daylight is the method, and Proofing on screen or on paper is which faults each medium will and will not show you.
What a proof cannot do is guarantee the manufacture of every later copy, because it is one copy, made once, on one machine. It is evidence about your file. It is a sample of what that file produces, not a specimen every future copy is matched against.
What legitimately varies
Ink density. A press drifts between calibrations and is brought back at service. A copy printed late in a maintenance cycle can run lighter or heavier than one printed the morning after an engineer left. This is the most visible difference and the one authors report most often, usually as "the cover is washed out". Dark, saturated covers show it worst, which is one more reason not to build a design that depends on an exact color — see Why your cover prints darker than it looked.
Paper batch. Paper is a manufactured product with its own tolerance. Shade moves — cream is a family of colors, not one — and so does bulk, the thickness per page. Spine width is computed from a nominal figure: KDP counts white at 0.002252 inches per page and cream at 0.0025. Those are averages the printer publishes, and a real block can land a little either side of the arithmetic. On a book of a few hundred pages the difference is small enough to be invisible until you stand two copies side by side and find one marginally fatter.
The trim. The blade cuts the book after it is bound, through a stack of glued paper, and that cut has tolerance. Margins move a little, and so does where the fold falls on a wrap. Why a printed cover never lands exactly where you put it is the detail; Measuring your proof with a ruler is how to put numbers on it.
The plant. This is the one nobody expects. Print on demand routes an order to whichever facility is nearest the buyer, so a reader in another country is served by a different building from the one that made your proof. Those plants run different machines of different ages, fed by different local paper suppliers. They work to the same specification, which is not the same as being clones of each other. How print-on-demand actually works covers the sequence; Printing in more than one country covers what changes when you switch on a second territory.
In the tools. The wrap is drawn to your chosen printer's own geometry — trim, bleed, safe area and a spine computed from the real page count — so the variation you see in the hand is the press, not the file it was given. Start your book → · about the tools
Where variation stops and a defect starts
Three tests, in this order.
Is it consistent across copies? A fault that appears in every copy you own is not the press. It is your file, and no amount of reporting will fix it. Look at the file first, every time.
Is it one copy out of several? Then it is manufacturing: a mis-fed sheet, a bad glue line, a cover bound crooked, a crushed corner. Report that copy with photographs, through the printer's own support channel. Ordering author copies covers the practical side of checking a box as it arrives.
Can you see it at arm's length? Hold the book the way a reader holds it, in ordinary room light. A fault visible there is a fault worth reporting. A difference you can only find by holding the cover edge-on under a lamp, or by laying two copies together and comparing, is manufacturing variance, and reporting it will not change what the next copy looks like.
That last test is the useful one, because the proof-reading habit trains you to inspect a book far more closely than anybody will ever read it.
Order one copy the way a reader does
Author copies come to you through the printer, on the account that publishes the book. That is a different route from the one a buyer uses, and it is the route you always see.
So once or twice a year, buy your own book through the retail store, at retail price, shipped normally. It is the only way to see the copy the retail channel actually produced: the plant that serves that market, the packaging it arrived in, and how the object compares with the listing that sold it. Do it for each territory that matters to you, and do it after any change to the interior — Fixing a book that is already published explains what a corrected upload does and does not reach.
Keep a dated copy from every run
Put the date and the source in pencil inside the back cover — proof, author copy, retail, and which month. Shelve them together.
Then the next time a copy looks wrong, you are not comparing it against memory. You are comparing it against the proof you approved and a copy from six months ago, and the question "has this drifted or was it always like this?" has an answer sitting on the shelf. Keeping a record of what you checked is the paperwork half of the same habit.
Make your book
The tools are free to use: name your book, bring your manuscript or your artwork, and work as long as you like. You only pay when you want the files without the preview watermark.
Next: Ordering author copies · More: Order the proof. Then read it in daylight · The proof came back wrong. Now what? · Comparing proofs from two printers