Why the datasheet accuracy isn’t the accuracy you’ll get
It was measured in lab conditions on a reference surface. Here are four things that separate it from your reality.
Every manufacturer publishes a linearity figure, and every such figure was measured under conditions written in a footnote nobody reads: constant temperature, a matt ceramic reference surface, a perpendicular angle, and averaging over a carefully chosen number of samples. Your part meets one of those four conditions, at best.
Temperature
The thermal drift coefficient is specified separately from linearity, and it is often the dominant factor. A sensor with 0.1% of range linearity and 0.02% per degree drift will lose its entire linearity advantage over a five-degree change in the production cell.
Target angle
Datasheet figures are measured at a perpendicular angle. A part presented at ten degrees shifts the reflection point, and in triangulation that is a systematic error, not noise — it won’t average away.
Averaging and the rate you actually sample at
Many resolution figures are specified at a low measuring rate, where the sensor averages over a long window. Run the same sensor at the rate your application needs and the noise grows, sometimes by an order of magnitude.
What to do instead
Define the tolerance your process has to hold, and ask for a measurement on your part under your conditions. That is what a feasibility test is, and that is why it’s worth a week.
