Crowd-source measurements
haveyaseen opened this issue · comments
Collect and analyze raw data in a testing phase to make risk calculations more precise.
- User activities (walking, shopping, talking)
- Phone location (pockets, out of pockets, in a room)
- RSSI
- Device model
- Percentage of infected users
- Total amount of users
What do you think? Expert opinions are highly appreciated.
Checking if someone is talking is very critical in my opinion: This should require the app to have access to the microphone and could make the app less trustworthy to a major part of the user base. If we are communicating that we only collect anonymized IDs via Bluetooth, why do we need access to the microphone?
There isn't necessarily a need for microphone access, and I would assume that determining if someone is talking is an entire project itself.
I've been thinking about it a little and I think the user should set this metadata themselves and later send the collected to the server, together with a few random values. This could even be part of the live app when using differential privacy (algorithms like RAPPOR, Encode-Shuffle-Analyze) to randomize the relevant fields.
Just some technical thoughts about how to integrate those capabilities into the codebase:
A key question is, wether or not additional measuring capabilities should all become part of the public app or if there will be different versions for different audiences (e.g. a closed testing audience). If there will be different versions, maybe those capabilities can be implemented using build-time feature toggles. Environment variables during the build process could switch certain features on or off. That would allow to build different versions of the app with different capabilities including a version with enhanced measuring for a closed test audience for example. A transparent/public build and deploy pipeline could show the feature toggles used during the build. The feature toggles could be documented (e.g. regarding their potential affection of privacy) to address concerns.