Comment by publlus_enigma

I wonder how this works in tandem with the Computational Image Pipeline?

From Apple: "Apple Reference Image provides users with an unalterable reference photo, visually confirming what the sensor saw at the moment of capture."

The pixels that arrive in the final image have already been significantly altered from what the sensor saw: demosaicing, noise reduction, Deep Fusion, artificial bokeh, etc. Is the signature anchored to raw sensor readout (making the reference a de-mosaiced digital negative, with all CIP processing deliberately left outside the trust boundary), or to the CIP's emitted image (in which case what does 'unalterable' actually protect against)? If it is the latter, then Apple are essentially saying some forms of computational (and ML) manipulation are acceptable in a signed capture, so long as it is their manipulation.