just presented our work at #ase2025 and noticed I didn't talk about it here, so here we go:

did you know that android has free aggregated app usage information available on Google play? Cloud Profiles are just that and we did the first empirical analysis on them, with an overview of what kind of profiles exist on Android.

We also introduce Profile Coverage, a metric to estimate how much interaction with an app matches that the aggregated data. To measure it, we also developed a uprobe-based method tracer that works on emulators and hardware devices, and doesn't require rebuilding the system or the app!

Evaluating Monkey, DroidBot, and Fastbot2, we were surprised the random interactions held up with model-based techniques and all tools only achieve only a Profile Coverage of 70% max. Compared to code coverage we also found cases where increasing the code coverage actually reduced the profile coverage - hashtag-futurework to improve this.

This was done with the awesome Felix Kehrer, Jürgen Cito, and Martina Lindorfer. Code's available and the paper is over here

lool, can't clone the Android Open Source Project (AOSP) with the default parallelization settings in the documentation because it's throwing "HTTP 429 - Too Many Requests" Errors. Probably because they have to rate-limit AI scrapers. Probably because they hype AI as the great thing it isn't.

also as an aside: the third step in setting up the AOSP is o download the proprietary binaries. I am not kidding: https://source.android.com/docs/setup/download