Ten thousand random configurations do not help when the signals inside a profile contradict one another. We examine how many real device configurations are useful and why rare hardware needs more care than ordinary hardware.
Competitors often compare the size of their fingerprint databases. Our measurements point elsewhere. Canvas and WebGL vary with the profile seed, so two profiles can differ even when they use the same base device configuration. The profile seed provides the variation; an exotic base device does not.
The number also raises a licensing distinction. The public dataset of 10,000 fingerprints is licensed under GPL-3.0, and NimbusX does not bundle it. Free users may download that dataset independently; paid plans use our own device library, captured with our collector and free of third-party dataset licensing.
A repeated configuration is not automatically the problem; an incoherent one is. A graphics card from one generation paired with a chip from another, too many fonts for the operating system or a display never sold with that hardware all create contradictions. A detector looks for those contradictions inside one profile, not just matches between accounts.
A pool of 100 to 300 common configurations from real devices is usually enough. Diversity then comes from profile-level variation. One rule matters more: uncommon hardware should not repeat across many profiles on the same platform.
One measurement puts the scale in context. Fonts carried the most entropy: 9.45 bits; the font list alone made 35% of machines unique. Device selection therefore follows what the host machine can actually expose, not a random draw. A persona that expects 368 fonts on a computer with 147 can expose only the overlap, not the claimed identity. Measured availability is 79% for personas with 201 fonts and 35–40% for those with more than 400.
Fifty accounts reporting an RTX 4090 and the same 386 fonts stand out more than fifty accounts using ordinary mass-market hardware. Rarity is itself a signal. NimbusX validates every configuration before it enters the library, marks uncommon devices in the list and lets the engine choose hardware for mobile identities.
A separate health metric tracks reuse. The command device coverage shows how many profiles share each device record. If the value is above one, those profiles become easier to link and the device library stops serving its purpose.
Create two profiles on the Free plan and run both through the fingerprint test and independent public detectors. Canvas and WebGL differ, the device signals remain coherent, and the raw values show that directly.
How fingerprint checks connect profiles · Mobile fingerprints without a phone