dBm distribution, not single average in device data

[mkwiecinski]

I think it would be wise to include more data on signal strength distribution. At least 10th, 25th, 50th, 75th and 90th percentiles. This would allow applying more complex, probability-weighted rules when assessing x-device contact.

[jasonbay]

RSSI fluctuates from moment to moment. Our sense is to just take the mode and bias toward the stronger signals. What do you mean by "x-device contact", please?

[mkwiecinski]

How many RSSI readings does OpenTrace use to estimate distance? If 90% of signals at 2m coming to a device X lies in range (-60;-80) with a median of -68, then a single signal of -75 may be also a valid, 2m-away contact point (in ~20% of cases). How valid? Well, I would leave it to another estimation engine (yes, AI), that could learn in time if those weaker signals can also result in getting infected and, possibly, provide some aggregate risk score. If not to users themselves, at least to epidemiologists after a user has opted in to some kind of data sharing. In the end, the <2m criteria does not have to last permanently, it can be more complex statistical model that is based on exact RSSI readings of past encounters with infected and non-infected people. So if you take any signal below -68 as invalid no matter what, this - due to inherent randomness in BT radio signals - is not the best strategy you could in the long run.

x-device = cross device, just refering to any contact between two devices