Color meter that lights the object with the screen

Because those flashes are the measurement. A camera cannot tell the color of an object from the color of the light on it — the two arrive multiplied together. Photographing the same scene with the screen dark and then lit gives two frames whose difference contains only what the screen contributed, and the room, being identical in both, subtracts away. Against known spectra the result shifts 4 to 19% across a dark room, tungsten, daylight and fluorescent light, while the raw camera values shift 50 to 77%.

Because how much screen light lands on the object depends on how far away it is and at what angle, and nobody holds a phone the same way twice. Dividing the sample by a white patch in the same frame turns the reading from "how much light came back" into "what fraction of the light came back", which is a property of the object rather than of your hand. Measured: moving the object between ten and fifteen centimeters then changes the answer by under 12%, where the raw brightness would have changed by more than half.

Hold the phone steadier, and if you can, measure somewhere the light is not changing. The problem is that the phone adjusted its own exposure between the dark frame and a lit one, which makes the subtraction a comparison of two different cameras. The tool asks the camera for manual exposure first; many Android cameras grant it, iOS browsers do not, and when it cannot be locked the tool watches a patch the screen cannot reach and throws the reading away if that patch moves. That is why you get a warning rather than a wrong number.

No, and in a specific way worth knowing. A laboratory instrument measures absolute color against a calibrated standard, in units another instrument would agree with. This measures reflectance relative to whatever white you put in the frame, with a camera whose three channels are broad and overlapping. What it is good at is comparison: whether two things measured the same way are the same, and by how much. That is most of what people actually need, and it is the part a phone can do honestly.

It can measure the pads reliably, which is the hard half. The other half is the chart: turning a color into a number needs the manufacturer’s scale, and those differ between brands and even between batches. The practical way to use this is to measure the pads on the strip AND the reference patches on the bottle’s chart in the same session, under the same screen, and let the comparison do the work. That is more trustworthy than any app that matches your photo against colors stored on a server.

It depends on the pair, and the honest answer is measured rather than assumed. For two colors that differ in brightness or in a smooth, gentle way, the colored lights add nothing — tested at a 0.95 ratio, which is to say no gain at all. For a metameric pair, two genuinely different spectra that a single white light renders identically, they doubled the separation. The colored flashes cost about a second, so they are worth taking; but the big win is the subtraction of the room, not the extra colors.

Because they are measuring different things. A photo gives you radiance: how much light arrived at the sensor, which depends on the object, the lamp, the distance and the camera’s own decisions. This gives reflectance: what fraction of the light the object returned, with the lamp and the distance divided out. A sample reading 0.62 is returning 62% of what the white paper returns, and that number stays the same in a different room — which is the entire reason for going to the trouble.

It is coarse rather than wrong. The room cancels exactly in the arithmetic, because it was identical in both frames; what is left is how finely the camera could record the screen’s contribution, which is a difference between two large numbers. Measured on known spectra: when the screen supplies about three quarters of the light the reading holds to about 3%, at a half to about 5%, at a third to about 8%, and below a quarter it drifts past 13% — which is where the page stops printing decimals as though nothing had happened and tells you instead. Moving the object closer, turning the brightness up, or shading it from the window all fix the same thing.