Read what the percentage refers to
A specification written as ±(p% of reading + n counts) has two separate terms. The percentage grows with the reading. The count term depends on the resolution of the selected display range, so it can remain significant when the reading is small. This calculator applies that particular additive form; a percentage of full scale, a calibration correction or a differently written specification needs its own interpretation.
For reading R in lx and resolution q in lx per count:
Allowance a = R × p/100 + n × q.
The displayed specification band is max(0, R − a) to R + a, in lx. Clipping the lower end at zero reflects the nonnegative physical quantity, without turning a negative mathematical endpoint into a claim of better accuracy. The inputs still need to describe the applicable instrument range and conditions.
An artificial range at two light levels
Suppose an invented meter range has a 5% reading term, two additional counts and 1 lx per count. These are arithmetic examples, not a specification for an actual instrument.
The fixed count term is a much larger share of the low reading. A different range may change both its resolution and its published tolerance. Do not carry the 1 lx resolution into another range simply because the instrument name is the same. Results use two decimal places, with a below-display-precision marker for tiny nonzero values; calculations use unrounded inputs.
A tolerance band is not a confidence interval
NIST’s Type B uncertainty guidance explains how specifications and calibration information can contribute to uncertainty evaluation, including the role of distributions. Adding the two specification terms here does not perform that evaluation and does not establish a 95% probability.
Cosine response, spectral mismatch, calibration, drift, sensor positioning, shadows and changing light can affect a measurement. Whether any of these are already included in a particular published specification depends on that document. Keep its conditions with the result rather than adding another percentage without understanding its basis.
When interpreting sample statistics, a larger number of readings does not automatically cancel a shared meter bias. A paired daylight factor also depends on two readings and their common conditions; this single-reading band does not propagate their uncertainty through the ratio.