Fingerprint entropy measures how much identifying information a signal or set of signals carries, expressed in bits. One bit of entropy halves the pool of possible matches; the more bits a fingerprint accumulates across independent signals, the smaller the fraction of the population that shares that fingerprint, and the more uniquely identifying it becomes.
A signal contributes entropy proportional to how many distinct values it takes across the population. A signal that takes only two equally common values contributes one bit; a signal with many rare values contributes many bits. The total entropy of a fingerprint is highest when its constituent signals are independent, because correlated signals overlap in the information they provide.
In practice, no set of signals is perfectly independent. Signals that come from the same hardware subsystem, such as GPU renderer and GPU capability limits, tend to be correlated. Selecting signals from diverse subsystems, such as graphics, audio, platform, and fonts, captures more independent information and yields a more distinctive fingerprint for the same number of collected values.
Frequently asked questions
How many bits of entropy does it take to uniquely identify a browser?
With roughly a billion active browsers in the world, approximately 30 bits of entropy is enough to narrow a fingerprint to a single browser on average, since 2 to the power of 30 is just over one billion. In practice, fingerprinting systems aim for more than this to account for population skew and correlated signals.
Why do fingerprinting systems use many signals instead of one?
No single signal is uniquely identifying on its own: screen resolution or time zone alone matches millions of users. Combining many moderately informative signals multiplies the entropy as long as the signals are independent, making the combined fingerprint far more distinctive than any individual signal.
Does blocking one fingerprinting signal significantly reduce entropy?
Blocking a single signal reduces the fingerprint entropy by at most that signal's contribution, but a well-designed fingerprint spreads entropy across many independent signals. Blocking one may have little overall impact if many others remain, and the act of blocking is itself a detectable pattern.
