The address is not an identity

A peripheral announces itself by broadcasting advertising packets on three channels. The packet carries a six-byte address, which would be a convenient identifier if it stayed still. It does not. Devices using resolvable private addresses rotate that value on an interval the manufacturer picks, and the specification suggests keeping it under fifteen minutes.

The practical effect is that the same pair of glasses can appear as several devices over an hour, and two devices can never be safely assumed to be different hardware. Any scanner that reports a count is reporting distinct advertising identities in a window, which is a different quantity from distinct objects in a room. We treat the two as separate numbers, and we would be suspicious of any tool in this space that does not.

Identification runs on an allowlist, and allowlists rot

What remains usable is the payload. Service UUIDs and manufacturer-specific data can distinguish a known product, which turns detection into pattern matching against a list somebody has to maintain by hand.

That list covers the recognisable end of the market well enough. Most coverage of the category tracks the same few names, as in HackerNoon's piece on the Samsung and Meta rivalry, and those are the devices whose signatures are easiest to pin down. The detection problem lives somewhere else, in the unbranded long tail sold under a dozen marketplace names, frequently sharing firmware, frequently advertising nothing that separates them from a pair of earbuds. Every one of those is a false negative until somebody sits down with a sniffer and adds it.

The failure gets worse when it matters most

Here is the part that inverts the intuition. Many peripherals stop advertising once a connection is established. A pair of glasses sitting idle in a bag can therefore be louder on the air than a pair actively paired and in use.

Signal strength does not rescue this either. A body between the two radios costs you more than several metres of open air, so proximity readings are soft in exactly the crowded rooms where the question comes up. Our position is that a tool built on this should never render a confident zero. Reporting nothing detected is honest. Reporting nobody nearby is a claim the radio cannot support, and building an interface that implies otherwise is the easiest way to turn a limited signal into false reassurance.