The programme where the instrument is caught, and then repaired.
Twice here a piece is followed by its ghost: the same render, made the same way, from the same cube after every value in it was shuffled into a random position. They are very hard to tell apart. That is the finding rather than a lapse of attention — measured, the feature driving pitch is statistically indistinguishable between the real cube and the shuffle.
Then the mapping is changed. Pitch is driven by edge activity, a channel that does survive shuffling, instead of by brightness, which does not. The ghost stops matching: the real cube’s line covers twice the pitch range of its control.
Near the end, two recordings with the view deliberately frozen, so you can hear how little the bands move on their own. That is what established the accurate version of the finding: the melody was reporting that the view changed, and could not report what changed.
This programme contains material from the DESI Early Data Release (DESI Collaboration et al. 2024), used under CC BY 4.0. Required citation and the full acknowledgement: https://radio.enthropysystems.com/pages/attribution