Every frame is read as raw sensor data, which is already linear, so the shutter speed, aperture and ISO in its metadata say exactly how much light each one caught. Putting them all on the same scale makes them the same picture, and each pixel is then taken from the frames that recorded it well: not so dark that it is mostly noise, not so bright that the sensor ran out. This is what Lightroom builds its 32 bit floating point DNG from, and like Lightroom the result is not a finished photograph, it is a wider negative to develop.