Because distance is not what a plant responds to. What matters at the leaf is photosynthetic photon flux density, the photons between 400 and 700 nanometres arriving on a square metre each second. Over a day that becomes the daily light integral, which the University of Kentucky's greenhouse horticulture library defines as moles of photons in the same waveband per square metre per day. The same note records what growers install: many ornamental growers deliver 60 to 100 µmol/m²/s at crop level, while growers of fruiting vegetables typically provide at least twice that intensity. Two panels drawing identical wattage can land on opposite sides of that range at the same height, because their lenses and reflectors spread the light differently. A height copied from another grower's tent is a guess about their hardware.

Real numbers do exist, as long as you read them as the consequence of a target rather than as a rule. The Google Play description of the Android build sets out the same order of operations:

A 2026 vertical-farm study of lettuce in BMC Plant Biology reports the mechanics plainly: the distance between the LED panels and the plant canopy was held at approximately 26 to 28 centimetres, varying with growth stage, in order to keep light intensity at roughly 200 µmol/m²/s of photosynthetic photon flux density at canopy level, measured with a portable spectral irradiance spectrometer. An 18 hour photoperiod turned that intensity into a daily light integral of about 13.0 mol/m²/d. Notice the sequence: the team picked the intensity, moved the panel until a meter agreed, then adjusted it as the plants grew. The 26 centimetres is an outcome, not an instruction.

Four steps, under ten minutes. Pick the plant preset first, so you have a target band rather than a vague sense of bright. Lay the phone where the topmost leaves sit, sensor upward, with the lamp at the setting you will use. Record an average rather than one flicker of a number, then move the lamp in five centimetre steps and read again; one step usually crosses more of the band than people expect. Convert to a daily total to check the photoperiod: intensity in µmol/m²/s, multiplied by hours of light, multiplied by 3600, divided by 1,000,000. For the lettuce trial that is 200 times 18 times 3600 divided by 1,000,000, or 12.96 mol/m²/d, which is the 13.0 the authors report.

Because the rule people reach for describes a lamp that does not exist. The inverse square relationship, where light falls off with the square of the distance, belongs to a point source. A panel or a bar array is a broad surface, so immediately beneath it the fall-off is gentler than the rule predicts, approaching it only well away from the fixture. Worse, a fixture is brightest under its centre and dimmest at its corners, so raising it trades peak intensity for evenness. A single reading in the middle of a tent therefore flatters the plants at the edges. Measure the centre, the four corners and one midpoint, and treat the lowest of those six numbers as the one your slowest plant lives with.

Well enough to tell 150 from 400, which is the decision you face. Android exposes the hardware: the SensorEvent reference states that for Sensor.TYPE_LIGHT, values[0] is the ambient light level in SI lux units. Lux is the problem, because the unit is built around human vision. NIST's definition of luminous intensity fixes the luminous efficacy of radiation at 540 times 10 to the 12th hertz at 683 lumens per watt, so lux weights green steeply and deep red barely at all. A meter must convert with a coefficient for your lamp. On iPhone the better route is the camera: AVCapturePhotoOutput supports RAW capture and DNG output, an exposure a converter can trust.

It is the wrong input when heat is the real constraint: a photon count cannot tell you a leaf is scorching under a diode; only a thermometer or your hand will. It ignores anything outside 400 to 700 nanometres, so a fixture sold on its far-red or ultraviolet is judged on part of its output. And without a calibrated camera profile the value is an ambient sensor estimate, which settles a 5 centimetre decision but is not publishable data.