What the sensor actually sees
Every setting you can change happens downstream of the optics. A scanner is a light source, a sensor that reads one thin line at a time, and something between them that either uses a lens or does not. Those choices decide how sharp a scan can possibly be, how tolerant the device is of paper that will not lie flat, and why one slip looks different on two machines at identical settings.
One line at a time
A scanner does not photograph a page. It reads a single row of the image, moves the paper or the sensor a small step, reads the next row, and assembles a file out of thousands of rows in sequence. That is why transport speed shows up as geometry rather than blur, and why a defect in the optical path repeats identically on every page of a batch — the same detectors read every row of every page. Reading a defect’s shape to find its cause is the subject of diagnosing scan defects by their shape.
The number of detectors across that row is the real limit on horizontal detail. Anything the driver offers above it is arrived at by calculation rather than measurement, which is one reason the numbers in what dpi does to a receipt scan do not all mean the same thing.
Two ways to build the gap between paper and sensor
Either you put the sensor almost against the paper, or you put a lens in the way.
A contact-type sensor sits directly behind the glass with its detectors spread across the full width of the scan line. No lens, no folded light path. It is compact, cheap, mechanically simple and needs no warm-up, which is why small portable and bus-powered devices are almost always built this way.
The alternative uses reduction optics: light from the full width of the page is folded through mirrors and focused by a lens onto a much smaller row of detectors. This is bulkier, needs alignment, and historically needed a moment to stabilise its lamp. What it buys is focal depth.
Depth of field is why a curled slip goes soft
A contact sensor is focused at the glass and almost nowhere else. With no lens the working distance is tiny by design, so anything not pressed flat against the surface is out of focus. A lens-based path has a focal depth measured in millimetres rather than fractions of one — which sounds trivial and is the whole difference in practice.
This explains a symptom that otherwise looks like a mystery. A receipt that lived on a roll keeps the curve of the roll. In a sheetfed path the transport holds it against the scan window; on a flatbed with the lid closed it is flattened. But a slip resting on an open flatbed, a slip sitting on a stack, a folded slip whose crease lifts it off the glass, or a stiff card that will not conform — these sit above the focal plane, and on a contact-type sensor a fraction of a millimetre of lift is enough to soften print that is already small and low in contrast.
The consequences are undramatic and worth doing. Flatten before capture, not after; nothing downstream recovers softness. Close the lid or weight the paper. Present creases flat rather than trusting the mechanism to press them out. And when choosing between the form factors in sheetfed, flatbed, or overhead, note that an overhead camera trades flatness for focus-at-distance rather than removing the problem.
The light source and what it does to faint print
The light source has to be even across the full width of the scan line, and unevenness there produces a soft gradient down the length of every page. It also has a colour: a source weighted differently across the spectrum renders the same aged thermal coating with more or less contrast against the paper, which is one reason two devices disagree about whether a slip is legible.
That interacts directly with fading. Thermal print loses contrast against its own background rather than losing shape, so how much contrast the illumination generates determines how much there is left to work with, before any exposure setting is applied. It is also why a threshold applied at capture can succeed on one machine and fail on another with the same paper.
The white strip you never see
Every scanner reads a known white reference before it reads your paper. A strip of white material sits inside the device, usually just outside the scan area or on the underside of the lid, and the machine scans it to learn what its own sensor and lamp currently do. From that it builds a per-detector correction: this detector reads a little dark, that one a little bright, the lamp is dimmer at one end. Every subsequent scan is adjusted by it.
This is why a dirty strip is a real fault with an odd signature. A mark on the white reference is read as “this detector is dark”, so the device brightens that column — in your scan, permanently, in the same place on every page, in the direction opposite to what you would guess. A stripe lighter than the page around it is a reference problem far more often than a glass problem.
Two things follow. Whatever cleaning routine you run should include the reference surface and not only the scan window, which is part of scanner cleaning and consumables. And if a device offers a calibration routine, running it after cleaning is worth the minute — it is how the device notices the surfaces changed. Lamps age too, and the correction is what hides that ageing until it can no longer compensate.
What this changes about your settings
Not much, deliberately, and that is the point. The optics set a ceiling; the settings decide how much of it you use. Knowing which build you have tells you where to spend effort: on a contact-type device flatness is the highest-value thing you control, and no exposure adjustment substitutes for it. On a lens-based device flatness matters less and the lamp’s condition matters more.
It also explains two devices with identical specifications producing visibly different files. The specification describes the sampling. The optics describe what was there to sample.
Whether a scan is adequate as a record is a separate question from whether it is sharp, and the rules differ by country and by entity type. Put that one to your own tax authority or an accountant in your jurisdiction.