Testing results

Per-device status

The question we are answering

Whether the glasses we study emit Bluetooth Low Energy signals that (a) exist in realistic states, (b) can be matched with at least two independent kinds of evidence, and (c) are not shared with non-glasses hardware. The outcome is a GO or NO-GO decision for the consumer app, made on evidence rather than hope.

Hardware priority

1. Ray-Ban Meta (Gen 1 or Gen 2): required.
2. Oakley Meta: if accessible.
3. Control hardware for false positives: a Meta Quest headset and controllers (same company identifier family), AirPods and other wearables, a second iPhone.
4. Later: Snap Spectacles, RayNeo, Rokid, Xiaomi AI glasses, Even Realities G1.

Protocol

Each scenario is one export, labelled with device, power state, case state, pairing state, activity, distance, environment and iOS state. We do not attempt the full cross product; we run the blocks below in order and stop early if the first block shows nothing is advertised.

Block A: does it advertise at all? (foreground, quiet room, 1 m)

• Glasses in closed case; case opened with glasses inside.
• Glasses out, unpaired, in pairing mode: name, service UUIDs, manufacturer payload.
• Paired to the test iPhone and connected, idle: do advertisements continue once connected?
• Paired but disconnected: the reconnect advertising pattern.
• Paired to a second phone, connected, idle: the bystander case, and the critical scenario.
• Paired to a second phone while taking a photo, recording video, playing music or using the voice assistant: any change in payload or rate.
• Glasses powered off: must be silent (a false-positive check).

Block B: RSSI against distance

0.5, 1, 2, 5 and 10 metres, 60 seconds each, phone held normally, in a quiet room and again in a busy environment. Recorded: strongest, smoothed and categorised RSSI. This is what tells us whether even coarse proximity categories are meaningful; see RSSI and distance.

Block C: iOS states

App backgrounded with the service filter off (expect nothing), backgrounded with the filter on (expect slower, de-duplicated observations), lock screen visible, screen off for two minutes then wake, and force quit with a restore identifier (expect no relaunch on iOS 26). Expectations come from Apple's documentation; the test checks them.

Block D: false positives with no glasses present

A Meta Quest with controllers at 2 m; a busy café for ten minutes; a normal daily environment for thirty minutes. We look for any MEDIUM or HIGH result without glasses present and record which evidence caused it.

GO criteria (all must hold)

• Glasses advertise something matchable in the bystander case, or at least in the unpaired, pairing-mode or disconnected states with a clear and honest product framing.
• At least two independent evidence kinds are available in the useful state, so MEDIUM is reachable without the single-dimension cap.
• The false-positive blocks produce no HIGH and only rare MEDIUM.
• Detection latency in the useful state is under 15 seconds in the foreground.

NO-GO or pivot triggers

• Nothing advertised once paired and connected: consumer value collapses to detecting glasses in pairing mode or in the case, and claims must be narrowed accordingly.
• Only a company identifier, with no service or payload: glasses cannot be separated from headsets, so LOW is the ceiling.
• Background scanning yields nothing even with the filter: no background feature.

What will be published here

• Date, iPhone model, iOS version and glasses firmware for every session.
• Per scenario: which advertisement fields were present, which evidence kinds matched, observation rate and latency, and the resulting confidence level.
• RSSI tables per distance and environment.
• False-positive findings with the exact evidence that caused them.
• The calibration decision and the resulting weights.
• The GO/NO-GO recommendation and its reasoning.

Results will be published whether they are favourable or not. If the answer is that a device cannot be detected in realistic conditions, its record in the device database will say so.

Related pages

• How Bluetooth detection works
• Where signatures come from
• False positives and false negatives