Cue everyone in tech arguing about if this includes any ML component's model weights (if it doesn't fail), and if it includes the data those weights were trained on (if it doesn't fail).

And if we did get that, then perhaps we get to follow up with all the people who keep demanding those things suddenly realising the state of the art in interpreting and debug weights is basically "doing animal psychology with an EEG and/or a TCMS".