Testing a scanner with your own paper

Every figure on a scanner’s box was measured under conditions chosen by the person selling it, on standard test sheets, at settings that maximise the number. None of that predicts what the device does with a handful of curled thermal slips from your wallet. The only test that answers your question uses your paper, and it takes about half an hour to set up once and a few minutes to repeat on any device thereafter.

Build a test batch, not a test page

Assemble twenty or so real items that between them cover your worst cases. The point is coverage, not representativeness — you already know the device handles a crisp sheet.

Include: the faintest slip you own, the longest, the most crumpled, the narrowest, one printed on tinted or shaded stock, one with handwriting on it, one that is torn, and a handful of ordinary ones for the baseline. Keep this batch. Its value is that it is the same paper every time, so results across devices and across months are actually comparable.

Also keep one clean printed reference page with fine text and a large flat area. That is your diagnostic sheet for hardware faults, used the way diagnosing scan defects describes, and it complements the receipt batch rather than replacing it.

What to measure

Five things, all countable, none requiring anyone else’s numbers.

First-attempt feed rate. Of the items you put in the tray, how many fed straight and singly without intervention? This is the measurement that best predicts whether you will keep using the device.
Multi-feed count. Items in versus files out. If the counts differ, the device is dropping documents silently, which is the most expensive failure mode in this category — the mechanism is in how a document feeder works.
Legibility of the hard items. For the faintest slip and the crumpled one: can you read the total, the date, and the vendor? A yes or no per item, not an impression.
Edge integrity. For the narrowest items, is anything clipped? Automatic cropping fails at the margins, silently, as deskew, autocrop and auto-rotate explains.
Hands-on time. Not the machine’s throughput — your involvement. Time yourself from picking up the stack to having usable files, including flattening, sorting, clearing a jam, and rescanning failures.

That last one is the number that matters and the one no specification reports. A device with an impressive rated speed and a poor first-attempt rate on your paper will consume more of your attention than a slower one that just works, because every intervention costs you a context switch rather than a few seconds.

The settings sweep

While you have the batch out, use it to settle your settings once rather than fiddling per batch forever.

Scan the faintest slip in each colour mode, and at the default resolution plus one step up and one step down. View them all at the same on-screen size and note which combinations keep the total readable. That single exercise usually establishes that greyscale at the default resolution is correct and that you can stop thinking about it — the reasoning behind each control is in colour, greyscale, or bitonal and what dpi does to a receipt scan.

Then do the same with the automatic cleanup features on and off, because their effect on faint print is device-specific and often surprising. Write down what you chose. You will not remember in six months, and the settings will get reset by a software update at some point.

Reading a specification sheet afterwards

Once you have your own measurements, the published figures become easier to interpret because you know what they omit.

A rated speed in pages per minute is measured on standard sheets at a specified colour mode and resolution, in continuous operation, excluding everything that happens before and after. It does not include flattening curl, sorting a mixed stack, clearing a misfeed, checking counts, or rescanning. On a receipt batch those dominate, so the rated figure and your experience will diverge by a wide margin — not because anyone lied, but because the number answers a different question.

The same applies to capacity figures, which assume uniform paper of a specified weight, and to resolution figures, which may be interpolated rather than optical. Treat all of them as upper bounds under favourable conditions.

Testing before you buy

The awkward part is that this method requires the device. Where you can, bring your batch to a shop and ask to run it, or borrow one, or buy from somewhere that takes returns and test in the first week. Where you cannot, at minimum test what you already have — the printer on your desk may pass, as the multifunction printer as a receipt scanner discusses — so that you know precisely what deficiency you are paying to fix.

For a used purchase the test is not optional, because condition is the entire question and it cannot be inferred from the model: buying a scanner second-hand.

Re-run it when things change

Keep the batch and re-run it in three situations. After any software update, because settings get silently restored to defaults. When jams or misfeeds start increasing, because comparing against your original first-attempt rate turns a vague impression into a measurement and tells you whether a part is worn — see scanner cleaning and consumables. And before buying a replacement, so the comparison is against real numbers rather than against your frustration.

The whole discipline here is refusing to accept anyone else’s measurement of your paper. Nobody publishing scanner figures has seen your receipts, and the specific properties that make receipts difficult — curl, coating, width variation, faded print — are exactly the ones standard test sheets are designed not to have.

What none of this settles is whether the resulting files are acceptable as records. Requirements for the form documents must take, and how long originals must be kept alongside them, vary by country and by the kind of entity keeping the records. That question belongs to your own tax authority or an accountant in your jurisdiction, and no amount of testing substitutes for asking it.