The warning in sleep is the silence between its signals

Sleep is a network, and the early signs of neurodegenerative disease appear to live in the moments the network falls out of step. The pattern that surfaced across roughly 124,000 overnight polysomnography recordings in the arXiv preprint on large-scale sleep physiology is that the warning of Parkinson's and Alzheimer's is network-level, not local. Heart, breathing, brain waves, and oxygen stop coordinating before they stop working, and that loss of coordination is what tracked future diagnosis.

Read the headline as "AI found a warning sign in your sleep" and you miss the actual finding. The arXiv preprint's hazard ratios of 1.48 for Parkinson's and 1.38 for Alzheimer's are population-scale, drawn from a clinical cohort rather than a wearable, and they apply to the way subsystems decouple, not to any single channel's value. The other findings carry the same shape: arousal architecture reorganizes the insomnia-apnoea boundary, REM duration follows the NREM that preceded it, and narcolepsy type 1 hides in a fast-sigma deficit. One mechanism, repeatedly.

The reusable move: a clinical-AI result is only as trustworthy as the code it ships with. The arXiv preprint links every reported number to executable code and routes every hypothesis through human review, which is the bar the next "AI sleep scientist" announcement will be measured against. Treating sleep as a network may enable more timely identification of disease risk than treating it as a list of channels.