Duplex scanning, and what blank-page removal deletes

Duplex means the device captures both sides of a sheet. It is sold as a single feature, but there are two quite different mechanisms behind the word, and they fail in different ways. The setting that usually accompanies it — automatic blank-page removal — is worth more attention than the duplex mode itself, because it is one of the very few functions on a scanner that deletes a whole page rather than degrading it.

Two mechanisms, one word

Some devices read both sides at once; others read one side twice. A dual-sensor design puts a sensor on each face of the paper path, so a sheet travels through once and both images are captured on the same pass. A single-sensor design has to get the paper past its one sensor twice — either by reversing the sheet back through the transport and flipping it, or by asking you to reload the stack face-down after the first pass.

The difference matters for receipts specifically, and not because of speed. A dual-sensor pass puts a slip through the feed path once. A reversing pass puts the same slip through the rollers a second time, in the opposite direction, after it has already been curled and gripped once. Every extra trip through a transport is another chance for the failures in why receipts jam to happen, and a short, curled, coated slip is the worst candidate for a reversal. A manual second pass has a different problem: the two images are now two separate files that you have to associate with each other by hand, and nothing in the software knows they belong together.

A third arrangement is a device with no duplex at all, which expects you to turn the paper over. For till receipts that is very often the honest answer.

Duplex on receipts is mostly a decision about the back

The reason duplex is close to mandatory on documents and optional on receipts is that documents are printed on both sides by default and slips generally are not. Turning duplex on for a receipt batch produces one image of the printed face and one image of a blank strip of thermal paper, for every item. That is a file roughly twice the size carrying no additional information.

But “usually blank” is not “always blank”, and the exceptions are exactly the ones you would not want to lose:

Card-terminal slips printing a signature line, a merchant copy notice, or the last digits of the card on the reverse.
Returns and exchange terms printed on the back, which is the one thing you might need to produce later.
Handwritten additions — a note about what the purchase was for, a project code, a name. On a slip with no useful description, that annotation may be the only thing making the record legible.
Continuation on a slip that ran out of roll and printed the remainder on the reverse.

So the sensible position is not “duplex off” but “duplex on, with something that discards the genuinely empty backs”. Which is where the trouble starts.

Blank-page removal is a threshold, not a judgement

The device does not know what a blank page is; it knows how much ink it measured. Blank-page detection works by computing how much of the image differs from the paper tone and comparing that to a threshold. Below the threshold, the page is discarded before it is written to the file. There is no model of meaning involved. A page with three faint characters in a corner and a page with nothing at all can land on the same side of the same number.

This makes it the most consequential setting on a receipt scanner, because its failure is total rather than partial. Aggressive cropping takes a column off an image, which is bad; aggressive blank-page removal takes the page out of the file, which is worse, and leaves no evidence that it did. You do not get a mostly-empty image to look at. You get a document with fewer pages than the paper had, and a page count that matches nothing.

One property of receipts makes the threshold much harder to set than it is on office paper: faint print sits close to the paper tone by definition, so a fading slip’s printed face can register as nearly empty — the same problem described in scanning faded thermal paper. Bitonal capture compounds it, because the grey values are gone before the detector ever sees the image.

A working policy for receipt batches

Prefer a device that flags removals rather than one that silently drops them. Some drivers report how many pages were discarded per batch. That single line of output turns an invisible failure into a visible one.
If the sensitivity is adjustable, move it towards keeping pages. A batch with some blank backs in it costs storage. A batch missing a signed card slip costs you the record.
Turn it off entirely for anything faint or old. Decide per batch if the device lets you, and default to off for the boxes of paper that have been sitting around longest.
Never combine aggressive blank-page removal with aggressive background removal. Background flattening pushes faint print towards the paper tone, and the blank detector then measures the flattened result. Together they can remove a page that either alone would have kept.
Check page counts, not images. The only cheap detection is arithmetic, and doing it deliberately is the subject of knowing you scanned every page.

Finding out what your device does

None of this is usually documented in terms you can act on, so you learn it the same way you learn every other default: feed the device paper you prepared and see what comes back. Build a small stack containing one slip with a single faint mark on the reverse, one genuinely blank reverse, and one card slip with a printed back; run it with removal on, again with it off, and compare page counts. That is a short version of testing a scanner with your own paper, and worth repeating after any driver update, because this is exactly the kind of default that gets reset.

Whether a scan of one face is sufficient as a record, and whether the reverse of a slip forms part of it, depends on rules that vary by country and by the kind of entity keeping the records. That is a question for your own tax authority or an accountant in your jurisdiction, not for the scanner’s settings dialog.