The current techniques pushes as many signals as possible to get a device fingerprint, which is then correlated with activity. It uses the pre-LLM machine learning methods to get such a picture. Sources of these signals are often abuses of protocols used to enhance UX across devices. Examples include: - Fonts installed ( https://gist.github.com/szepeviktor/d28dfcfc889fe61763f3 ) - Power saving telemetry ( https://dev…
Aggressive DNS blocking gets much better results. Bromite / similar beowser also helps.