You tap the shutter. The phone chimes, or vibrates, or does nothing at all, and a photo appears in your camera roll looking perfectly fine. What you don't see is the half-second of quiet demolition that happened in between, where your phone took everything its sensor just recorded and threw a significant chunk of it away.
Not a glitch. The design.
What the sensor actually sees
The raw data coming off a modern smartphone sensor is enormous. A 50-megapixel sensor reading 12 bits of color depth per channel produces a file north of 100MB for a single frame. Your phone takes that, chews on it for a fraction of a second, and hands you something closer to 4-8MB. So what happened?
The sensor itself doesn't see color. It sees light intensity. Each pixel sits behind a colored filter, red, green, or blue, arranged in a checkerboard pattern called a Bayer array (named for Kodak engineer Bryce Bayer, who patented it in 1976). About half the pixels are filtered green, a quarter red, a quarter blue, because human vision is most sensitive to green. The chip reads raw brightness values from each filtered pixel and nothing else. No color, no sharpness, no exposure decisions. Just 50 million numbers.
Turning that into something recognizable requires demosaicing: the phone's image signal processor interpolates the missing color values for each pixel by sampling its neighbors. A red pixel borrows green and blue estimates from the pixels around it. Do this 50 million times and you have a full-color image, but now you're carrying three values per pixel instead of one, and the file is even larger than before.
Then the processor keeps going. Noise reduction, white balance correction, sharpening, tone mapping, HDR blending if the phone shot multiple exposures. Each step is a judgment call baked into firmware, and each one trades raw information for something that looks good on a phone screen.
The compression that finishes the job
After all that processing, the image gets encoded as a JPEG, which uses lossy compression. JPEG divides the image into 8x8 pixel blocks, runs a mathematical transform that separates detail humans notice from detail they don't, then discards the less-noticed parts more aggressively the harder you compress. Think of it like a hotel that strips the minibar and the extra pillows before you check in, then tells you the room is ready. At a typical smartphone quality setting, another 10-20% of the remaining image information is gone.
Here's a scenario worth sitting with. Two people buy the same phone on the same day: Priya and Tom. Priya shoots in default auto mode and shares photos straight to a messaging app, which re-compresses them on upload. Tom enables the RAW option in a third-party app, gets the unprocessed Bayer data as a DNG file, and edits in Lightroom. Priya's final image has passed through the ISP pipeline, JPEG compression, and then the messaging platform's own encoder. Three rounds of lossy compression, total. Tom's DNG is still carrying the original 12-bit sensor data. When he pulls the shadows in post, smooth gradients appear. When Priya tries the same edit on her JPEG, blocky artifacts fill the dark areas, because the data that would have saved those shadows was discarded at step one.
That difference is real. It is not audiophile-level placebo. It's why professional mobile photographers shoot RAW even when it means managing 80MB files.
One thing people consistently get wrong
Most people assume computational photography features like Night Mode or Portrait Mode preserve more information because they feel more sophisticated. Often the opposite is true.
Night Mode on most phones stacks anywhere from four to thirty frames, aligns them, merges them to reduce noise, and then flattens the whole thing into a single processed JPEG. You get a cleaner-looking image and a smaller file, but the underlying dynamic range of those thirty raw frames is gone. The phone made all the decisions for you. Sometimes that's exactly right. Sometimes you needed those decisions to be yours.
Ask yourself honestly: how many photos have you tried to rescue in editing, only to find the shadows had already turned to mush?
The image signal processor is not your enemy. For the vast majority of shots, the pipeline it runs produces images that look better to the human eye than unprocessed sensor data would. Raw Bayer data looks flat, greenish, and noisy until someone processes it. The processor does that work invisibly and instantly, and it does it well.
But the compression is irreversible, and that matters. Every JPEG you save is a closed door. The phone decided what mattered in that scene, locked it in, and deleted the rest. Knowing that doesn't mean you should shoot RAW constantly. It means the people who understand this pipeline get to make a choice, and everyone else just gets the phone's opinion of their photograph.