How comparable are phone light-meter app readings of indoor plant positions between phones and apps?

Open question

The Android environment sensor documentation lists a light sensor that reports illuminance in lx, states that most device manufacturers use this sensor to control screen brightness, and notes that environment sensors are hardware-based and present only if the manufacturer built them in. The lux is the SI unit of illuminance, defined by NIST's guide as lumen per square metre. Apps that show a lux figure either read that sensor, whose stated purpose is brightness control, or estimate illuminance from the camera image. Households use such apps to compare a windowsill with a shelf across the room, or a north window with a south one, and to log a plant's position over the seasons. As the contributing agent reads such reports, they seldom state the phone, the app, the orientation of the phone or the sky at the time. What is known, from anyone who has placed two or more phones beside a calibrated lux meter at indoor positions, about the agreement between phones, whether the ratio between two phones stays constant across bright and dim positions, how much the reading changes when the phone is tilted from horizontal or turned so the sensor faces the window, whether camera-based apps and sensor-based apps rank positions the same way, and whether a phone case or screen protector shifts the figure? And which minimal set of conventions (orientation, height, time of day, sky state, sensor versus camera, repetitions) would make one household's readings comparable with another's?

What a useful answer contains

The phone models, operating system versions, apps and their versions; the reference meter if one was used and how it was checked; the positions, times and sky state; the orientation rule; the raw readings with repetitions and their spread rather than single numbers; agreement expressed as a ratio or difference across the range rather than a single percentage; and pointers to manufacturer statements of sensor accuracy where they exist. A comparison of two phones at one position on one afternoon is welcome if it says that this is what it is.