Reported pressure levels differ from how many a person can actually control
Aliases: pressure steps · 4096-level pressure · force resolution · force JND
What it is
Marketing copy often lists 2048 or 4096 pressure levels. That integer is how many codes the digitizer ADC or protocol can encode. How many force steps a hand can reproduce, and an eye can tell apart in the ink, is a coarser set of controllable levels. The two rulers are not interchangeable: bit depth describes sensing and transport, not the force ladder a writer can use.
Why it happens
Force travels through mechanical hysteresis, sensor noise, quantization, filtering, and the brush map before it becomes pixels. A nib spring or elastomer reads differently on the upstroke than the downstroke. An ADC slices continuous force into codes; if the noise floor already spans several codes, those steps are jitter, not information. Motor control of grip force is limited by muscle spindles and cutaneous receptors; people hold far fewer stable force steps than the protocol offers, and they are better at relative change (“a bit lighter”) than at matching an absolute newton value. The ink side compresses again: if width only moves between 2 and 8 pixels, four thousand force codes cannot become four thousand visible widths. What matters is how many just noticeable force differences fit in the force band people actually use, and whether those steps repeat during press–release—not the integer on the spec sheet.
Studying it
Measure force difference thresholds with constant stimuli or a staircase, logging both pen reports and an external force sensor. Ask people to redraw a line “one step thicker than last time” and see whether force deltas and width deltas line up.
Independent variables: quantization depth (or deliberate downsampling), filter window, nib stiffness, on-screen width or opacity range. Dependent measures: repeatable force steps, “changed / did not change” misjudgements, contour jitter from noise, and the fraction of codes eaten by dead zone or saturation.
Spec numbers overstate the channel. Histogram the reported values: if real writing occupies a narrow band of codes, extra bits are unused.
Where it stops holding
Firmware calibration, temperature, and worn nibs drift; factory 4096 is not the effective ladder a year later. A game that treats “hard press” as a skill trigger needs two or three bins, not the full ADC. Illustration that wants a smooth taper will show steps or jitter if usable levels are few. Children, low-grip users, and gloved writing shrink the usable force band further.
Applying it
- Separate “has pressure” from “how many distinguishable thicknesses or densities” in product language; do not treat protocol bits as the experience metric.
- Spend mapping resolution in the force band people actually use; treat noise-dominated lows and early saturation as dead zone or soft clip, not as fake extra steps.
- Apply light temporal filtering to kill jitter without flattening the peaks of a press–release.
- How to check: ten lines, each meant to be slightly thicker than the last; count how many steps stay stably distinct, and whether the extremes hit dead zone or saturation. Tiny irregular bulges on the contour are usually quantization noise, not style.