move the lamp to
210 mm
from the card, measured screen face to paper. Tap anywhere when it is there.
This screen is the light source. It walks a fixed sequence: at each of ten distances it shows the emitter fully dark, then half covered, then fully covered, and at three of those distances it steps through a nine rung coverage ladder instead. The striped band along the top carries the step number, so the camera on the other device knows what it is looking at without the two of them sharing a clock.
Set the brightness of this screen to full, once, and then do not touch it. Full, rather than comfortable, for two reasons: a panel dimmed below full is usually chopping its light in time, which a short exposure then samples a random slice of; and the brighter the lamp, the further out the card still clears the signal floor. Point the other device's camera at a sheet of white paper, put this screen at the distance it asks for, and press measure over there.
Do not set the radius small. The whole point is the departure from an inverse square, and over the sweep this page walks, 90 mm out to 600 mm, that departure is 0.048 of an exponent at a radius of 30 mm, 0.022 at 20 mm and 0.005 at 10 mm. The instrument's own interval on its shipped anchor is about 0.018. So below roughly 20 mm there is nothing left to measure: the prediction and a plain inverse square have become the same curve, and agreeing with it would mean nothing. Bigger is better here, and the lamp is not obliged to fit on a phone.
idle
Hold a ruler against the glass and adjust pixels per mm until the scale below reads its true length. The radius you set here is the number to put into the instrument page's radius slider, and it is the only thing about this lamp the estimator is never told. If the instrument refuses with a complaint about the response curve, drag band brightness until the striped band looks about as bright as the lit paper: the band is the exposure reference, and it has to sit inside the range of values the paper covers.